Notes on Lie Groups

Zuoqin Wang
December 14, 2009
Contents
1 Introduction 5
1.1 Group as Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Course Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Topological Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
I Part I: Geometry of Manifolds 10
2 Smooth Manifolds 11
2.1 Denition of manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Tangent vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Vector Fields 15
3.1 Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Integral Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Dynamics of Vector Fields . . . . . . . . . . . . . . . . . . . . . . . . 18
4 Submanifolds 20
4.1 Submersions and Immersions . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Vector Fields on Submanifolds . . . . . . . . . . . . . . . . . . . . . . 24
5 The Frobenius Theorem 25
5.1 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 The Frobenius Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 26
1
II Part II: Basic Lie Theory 30
6 Lie Groups and their Lie Algebras 31
6.1 Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.2 Lie Algebras Associated to Lie Groups . . . . . . . . . . . . . . . . . 33
6.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
7 The exponential map 36
7.1 One Parameter Subgroups . . . . . . . . . . . . . . . . . . . . . . . . 36
7.2 The Exponential Map . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.3 Dierent Descriptions of the Lie Bracket . . . . . . . . . . . . . . . . 38
8 Linear Lie Groups 41
8.1 The General Linear Group . . . . . . . . . . . . . . . . . . . . . . . . 41
8.2 The Special Linear Groups . . . . . . . . . . . . . . . . . . . . . . . . 42
8.3 The Orthogonal Groups and Unitary Groups . . . . . . . . . . . . . . 42
8.4 Symplectic Groups and Other Linear Lie Groups . . . . . . . . . . . . 44
9 The Baker-Campbell-Hausdor Formula 46
9.1 Taylor Series Expansion on Lie Groups . . . . . . . . . . . . . . . . . 46
9.2 The Derivative of the Exponential Map . . . . . . . . . . . . . . . . . 48
9.3 The Baker-Campbell-Hausdor Formula . . . . . . . . . . . . . . . . 49
10 Lie Subgroups 51
10.1 Lie Subgroups v.s. Lie Subalgebras . . . . . . . . . . . . . . . . . . . 51
10.2 Closed Lie Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
11 The The Lie Group-Lie Algebra Correspondences 55
11.1 Simply Connected Lie Groups . . . . . . . . . . . . . . . . . . . . . . 55
11.2 Lie Group Homomorphism v.s. Lie Algebra Homomorphism . . . . . 56
11.3 Lies Fundamental Theorems . . . . . . . . . . . . . . . . . . . . . . . 57
III Part III: Lie Group Actions 60
12 Lie Group Actions 61
12.1 Smooth Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2
12.2 Innitesimal Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
13 Proper Actions of Lie Groups 65
13.1 Orbits and Stabilizers . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
13.2 Proper Actions of Lie Groups . . . . . . . . . . . . . . . . . . . . . . 66
13.3 Proper Free Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
14 The Slice Theorem 70
14.1 Associated Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
14.2 The Slice Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
14.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
15 Orbit Types 75
15.1 Homogeneous Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
15.2 Orbit Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
IV Part IV: Representations of Lie Groups 79
16 Elementary Representation Theory 80
16.1 Representations of Lie Groups and Lie Algebras . . . . . . . . . . . . 80
16.2 Operations on Representations . . . . . . . . . . . . . . . . . . . . . . 81
16.3 Irreducible Representations . . . . . . . . . . . . . . . . . . . . . . . 83
17 The Haar Measure 85
17.1 Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
17.2 The Haar Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
17.3 Modular Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
18 Schurs Orthogonality 90
18.1 Schurs Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
18.2 Schur Orthogonality for Matrix Coecients . . . . . . . . . . . . . . 91
18.3 Schur Orthogonality for Characters . . . . . . . . . . . . . . . . . . . 93
19 The Peter-Weyl Theorem 95
19.1 Some Functional Analysis . . . . . . . . . . . . . . . . . . . . . . . . 95
19.2 Convolution on Compact Lie Groups . . . . . . . . . . . . . . . . . . 96
3
19.3 The Peter-Weyl Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 98
20 Applications of the Peter-Weyl Theorem 100
20.1 Faithful Representations . . . . . . . . . . . . . . . . . . . . . . . . . 100
20.2 Class Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
20.3 Irreducible Representation of SU(2) . . . . . . . . . . . . . . . . . . . 102
21 Abelian Lie Groups: Structure and Representations 105
21.1 Structure of Abelian Lie Groups . . . . . . . . . . . . . . . . . . . . . 105
21.2 Representation of the Torus . . . . . . . . . . . . . . . . . . . . . . . 106
21.3 Fourier Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
V Part V: Structure of Compact Lie Groups 110
22 Maximal Tori 111
22.1 Maximal Tori . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
22.2 Cartan Subalgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
23 Cartans Theorem 115
23.1 Surjectivity of the exponential map . . . . . . . . . . . . . . . . . . . 115
23.2 Cartans Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
23.3 Applications to Centralizers . . . . . . . . . . . . . . . . . . . . . . . 118
24 The Weyl Groups 119
24.1 The Weyl Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
24.2 The Maximal Tori of Classical Groups . . . . . . . . . . . . . . . . . 120
24.3 The Weyl Groups and Root Systems of Classical Groups . . . . . . . 122
25 Structure of Compact Lie Groups 124
25.1 Structure of Compact Lie Algebras . . . . . . . . . . . . . . . . . . . 124
25.2 The Commutator Subgroup . . . . . . . . . . . . . . . . . . . . . . . 125
25.3 Structure of Compact Lie Groups . . . . . . . . . . . . . . . . . . . . 127
26 The Weyl Integration Formula 128
26.1 The quotient G/T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
26.2 The Weyl Integration Formula . . . . . . . . . . . . . . . . . . . . . . 129
4
1 Introduction
1.1 Group as Symmetry
A group is an algebraic structure consisting of a set together with an operation
that satisfy group axioms. Groups are the mathematical way to describe symmetries.
To illustrate the power of symmetry in solving mathematical problems, lets
start by considering the following baby example.
Example: (see A. Kirillov, Introduction to Lie Groups and Lie Algebras.)
Suppose you have n (n > 2) numbers a
1
, , a
n
arranged on a circle.
You have a transformation A which replaces each number with the average
of its neighborhoods, i.e. replaces a
1
with
an+a
2
2
, a
2
with
a
1
+a
3
2
, and so on.
If you do this suciently many times, will the numbers be roughly equal?
To answer this, we need to study the asymptotic behavior of A
n
for
n large. Alternately, it suces to nd the eigenvalues and the eigenvec-
tors of A. While it is very hard to compute them from the characteristic
polynomial, we can actually compute them via symmetry. Note that the
problem has a rotational Z
n
-symmetry: if we denote by B the operator of
rotating the circle by
2
n
, i.e. sending (a
1
, a
2
, , a
n
) to (a
2
, a
3
, , a
n
, a
1
),
then BAB
1
= A. (This is Z
n
-symmetry since the operator B generates
the group of cyclic permutation Z
n
.) We will use this symmetry to nd the
eigenvectors and eigenvalues of A.
We will make use of the following basic result from linear algebra:
if two matrices A and B commutes, and V

is an eigenspace of B, then
AV

. (Check it if you never see it before!) In particular, A preserves

the eigenspace decomposition of B. So to diagonalize A, we only need to
diagonalize A on each eigenspace of B.
Back to our example. Since B
n
= I
n
the identical matrix, all the
eigenvalues of B must be n-th root of unity. Denote by = e
2i/n
the
primitive n-th root of unity. It turns out that the eigenvalues of B are
exactly
k
=
k
for k = 0, 1, , n 1, with corresponding eigenvectors
v
k
= (1,
k
,
2k
, ,
(n1)k
). Since each eigenspace of B is one dimensional,
each v
k
must also be an eigenvector of A! It follows that the eigenvalues of
A are
k
=

k
+
k
2
= cos
2k
n
.
Now we can answer the problem. If n is odd, then all eigenvalue except
the rst
0
(=1) satises [[ < 1, so that A
n
a will tends to an eigenvector
of
0
= 1, i.e. to (

a
i
n
, ,

a
i
n
). However, when n is even, we will have
another eigenvector
n/2
= 1, and so the numbers are not going to be
roughly equal, but oscillating between two set of numbers.
5
In this example, we found that the problem has a Z
n
-symmetry. Instead of
work on Z
n
itself, we realize the symmetry using the matrix B, or more precisely,
the cyclic group I, B, B
2
, , B
n1
. This is the idea of representation trying to
understand a group via its linear action on a linear space.
It turns out that the same idea works in many many other places. For example,
consider the sphere S
2
R
3
. Let be the Laplace operator on S
2
. It is very
important in both mathematics and physics to compute the eigenvalues of . Since
is a second order dierential operator, it is very hard to compute the eigenvalues
and eigenfunctions by solving the corresponding second order dierential equation.
But much as in the baby example above, it is easy to notice that this problem has a
SO(3)-symmetry. The only dierence is that now we have a very big group SO(3)
instead of the nite group Z
n
. The group SO(3) has very nice structure, and is a Lie
group. To nd the eigenvalues and eigenfunctions of , we need to study irreducible
representations of SO(3).
1.2 Course Plan
We have learned many groups from a standard algebra course. Typical ones
include Z, Z/nZ, Q as well as R, C, S
1
etc. In comparison with the rst there
ones, the last three groups have very nice geometric structure. They are Lie groups.
In short, Lie groups are groups with nice geometric structure such that the group
operations are compatible with the underlying geometric structure.
We have seen in some sense that representation theory will play an important role
in Lie theory. Another very natural and important tool to study Lie groups are the
associated Lie algebras: vaguely speaking they are linearization of Lie groups, just
as the tangent space can be viewed as the linearization of a surface. The advantage
here is that, according to the group law, every point looks the same (think of R, C
and S
1
), so that we only need to study the tangent space at one point. The simplest
choice is of course the tangent space of the identity element.
In this course we would like to cover:
Basic theory of manifolds and vector elds the geometry of the underlying
spaces of Lie groups and Lie algebras.
Lie groups and their associated Lie algebras, with emphasis on the use of the
exponential map.
Introduction to Lie group actions on smooth manifolds.
Basic representation theory and and role in the harmonic analysis on a Lie
group.
6
The structure of compact Lie groups.
1.3 Topological Groups
Denition 1.1. A topological group is a group G with the structure of Hausdor
topological space such that the maps
GG G, (g, h) g h
and
G G, g g
1
are both continuous.
Let G be a topological group G. For any a G, there is a natural left translation
L
a
: G G, g a g
and a right translation
R
a
: G G, g g a,
both are topological group isomorphisms, i.e. isomorphisms as group maps as well as
homeomorphisms between topological spaces. For any g G, L
g
1 as well as R
g
1
translate a neighborhood of g to a neighborhood of e. In other words, a topological
group is an object that the neighborhood of every point looks the same. Moreover,
Proposition 1.2. If G is a connected topological group and U is any open neigh-
borhood of the identity e G. Then G =

n=1
U
n
.
Proof. Since left translations are homeomorphisms, each U
n
is open. If follows that

n=1
U
n
is open in G. On the other hand, it is also closed, since if g ,

n=1
U
n
,
then the open set L
g
U
1

n=1
U
n
= . (We used the fact that U
1
is also an open
neighborhood of e.) Since G is connected, we conclude G =

n=1
U
n
.
A topological group is compact if the underlying topological space is compact.
Examples:
(1) S
1
is a compact topological group, while R is a noncompact topological Lie
group. (Nevertheless, R is locally compact.)
(2) Any group can be made into a topological group by endowing the underlying
space the discrete topology. In this case a group is compact if and only if it is a
nite group.
7
Compact topological groups are of particular interesting, partially because of
the existence of an invariant integral on G. Recall that an integral on G is a linear
map
I : C(G) R, f I(f) =
_
G
f(g)dg
such that f
1
(x) f
2
(x) for all x implies I(f
1
) I(f
2
). An integral is called left-
invariant if for all h G,
_
G
f(h g)dg =
_
G
f(g)dg.
We will always normalize the integral such that I(1) = 1.
Examples:
(1) If G is a nite group with discrete topology, then the invariant integral is
just the averaging operation
I(f) =
1
[G[

gG
f(g).
(2) If G = S
1
, then the following integral is left-invariant:
I(f) =
1
2
_
2
0
f(e
i
)d.
In fact, they are the only invariant integrals. To see this, for the case G is nite,
let
M(f) =
1
[G[

gG
f(g),
then by linearity and invariance, I(f) = I(M(f)). However, M(f) is a constant
function on G, so I(f) = M(f).
Similarly in the case of S
1
, for each m let M
m
(f) be the function
M
m
(f)(g) =
1
m
m

k=1
f(e
2ki/m
g),
then I(f) = I(M
m
(f)). Although M
m
(f) are no longer constant functions on G,
their limit as m is:
lim
m
M
m
(f) =
1
2
_
2
0
f(e
i
)d.
Under the same spirit, (but technically much more complicated), one can prove
8
Theorem 1.3. If G is a compact topological group, then there exists a unique nor-
malized left invariant integral.
We will not prove the theorem in this generality here, but will prove it for Lie
groups later.
9
Part I
Part I: Geometry of Manifolds
10
2 Smooth Manifolds
2.1 Denition of manifolds
Let U R
n
be open. Recall that a function f : U R is called smooth, denoted
by f C

(U), if for any multi-index = (

1
, ,
n
), the mixed partial derivative

f =

f
x

1
x

1
1


n
x
n
n
f
always exists. Similarly, we say that a map
f = (f
1
, , f
m
) : U V R
m
is a smooth map, if each f
i
C

(U) for all i.

A smooth map f : U V is called a dieomorphism if f is bijective, and
f
1
: V U is also smooth. It is easy to see that composition of smooth maps is still
smooth, and composition of dieomorphisms is still a dieomorphism. Moreover,
dieomorphism is a equivalent relation.
Denition 2.1. Let M be a (Hausdor and second countable) topological space.
It is said to be a n-dimensional topological manifold if for every p M, there exists
a triple , U, V , where U is an open neighborhood of p in M, V an open subset
of R
n
, and : U V a homeomorphism. Such a triple is called a chart about p.
Two charts (
1
, U
1
, V
1
) and (
2
, U
2
, V
2
) are called compatible if the transition
map

12
=
2

1
1
:
1
(U
1
U
2
)
2
(U
1
U
2
)
is a dieomorphism. Note that both
1
(U
1
U
2
) and
2
(U
1
U
2
) are subsets of R
n
.
Denition 2.2. An atlas / on M is a collection of charts

, U

, V

such that
all charts in / are compatible to each other, and satises

= M. Two atlas
on M are said to be equivalent if there union is still an atlas on M.
Denition 2.3. An n-dimensional smooth manifold is an n-dimensional topological
manifold M equipped with an equivalence class of atlas. This equivalence class is
called its smooth structure.
Some examples of smooth manifolds (with the god-given smooth structures):
R
n
, or more generally, any nite dimensional vector space, are smooth
manifolds.
Open subsets of a smooth manifold are smooth manifolds.
The circle S
1
is a smooth manifold.
If M and N are manifolds, so is their product M N.
11
Example: Smooth structure on S
n
R
n+1
: Recall
S
n
= (x
1
, , x
n+1
) R
n+1
[ x
2
1
+ +x
2
n+1
= 1.
Let U
1
= S
n
(0, , 0, 1) and U
2
= S
n
(0, . . . , 0, 1). Consider the stereo-
graphic projection maps

1
: U
1
R
n
,
1
(x) =
1
1 x
n+1
(x
1
, , x
n
)
and similarly

2
: U
2
R
n
,
1
(x) =
1
1 +x
n+1
(x
1
, , x
n
).
Obviously they dene charts on S
n
, with transition map

12
: R
n
0 R
n
0, (y
1
, , y
n
)
1
y
2
1
+ +y
2
n
(y
1
, , y
n
)
a dieomorphism.
One can dene smooth maps between smooth manifolds using their smooth
charts:
Denition 2.4. A map f : M N between smooth manifolds is called smooth if
for any chart

, U

, V

of M and any chart

, X

, Y

of N, the map

f
1

(U

f
1
(X

))

(f(U

) X

)
is smooth.
We will denote the set of all real-valued smooth functions on M by C

(M).
Any smooth map f : M N induces a pull-back map
f

: C

(N) C

(M), g g f.
Denition 2.5. We say that f : M N is a dieomorphism if it is bijective, and
that both f and f
1
are smooth maps.
Obviously if M is a smooth manifold, then any chart (, U, V ) gives a dieo-
morphism : U V from U M to V R
n
. Moreover, one can easily check that
two atlas / =

, U

, V

and B =

, U

, V

of M are equivalent if and only if

the identity map Id : M M is a dieomorphism.
Example: We can dene two atlas on R by / = id, R
(1)
, R and B = , R
(2)
, R,
where (x) = x
3
. They are non-equivalent atlas since the identity map Id : R
(1)

R
(2)
is not a dieomorphism. However, the map : R
(1)
R
(2)
, (x) = x
1/3
is a
dieomorphism. So the two inequivalent atlas are dieomorphic to each other.
12
Remarks (on smooth structures).
1. Its possible that a topological manifold supports many dierent(=non-
dieomorphic) smooth structures. In fact, a remarkable result of J. Milnor and
M. Kervaire asserts that the topological 7-sphere admits exactly 28 dierent smooth
structures! However, on any Lie group there is only one smooth structure.
2. The famous Whitney embedding theorem claims that any n-dimensional
smooth manifold can be smoothly embedded into R
2n+1
. In other words, any manifold
is a submanifold of R
m
.
2.2 Tangent vectors
Let M be an n-dimensional smooth manifold.
Denition 2.6. A tangent vector at a point p M is a linear map X
p
: C

(M) R
satisfying the Leibnitz law
X
p
(fg) = f(p)X
p
(g) +X
p
(f)g(p) (2.1)
It is easy to see that the set of all tangent vectors of M at p is a vector space.
We will call this the tangent space of M at p, and denote it by T
p
M.
Remark. There is a more geometrical way to dene tangent vectors: Take a chart
, U, V around p. Let (
p
be the set of all smooth maps : (, ) M such that
(0) = p. We dene an equivalence relation on (
p
by

dt
(0) =
d

dt
(0).
Then T
p
M = (
p
/ .
Theorem 2.7. T
p
M is an n-dimensional vector space.
Proof. Take a chart , U, V around p with (p) = 0. It is easy to check that the
maps

i
: C

(U) R, f
f
1
x
i
(0), i = 1, 2, , n
are tangent vectors at p. We claim that they form a basis of T
p
M. in other words,
T
p
M = span
R

1
, ,
n
.
The fact that
i
s are linearly independent follows from the identities

i
(x
j
) =
ij
.
13
On the other hand, for any f C

(U), the Taylors expansion reads

f(q) = f(p) +

(x
i
((q)) x
i
((p)))f
i
((q)),
where f
i
C

(V ) satises f
i
((p)) =
i
(f). Thus for any X
p
T
p
M,
X
p
(f) =

X
p
(x
i
)f
i
((p)) =

i
X
p
(x
i
)
i
(f).
In other words, X
p
=

i
X
p
(x
i
)
i
. So these
i
form a basis of T
p
M.
Denition 2.8. Let f : M N be a smooth map. Then for each p M, the
dierential of f is the linear map df
p
: T
p
M T
f(p)
N dened by
df
p
(X
p
)(g) = X
p
(g f)
for all X
p
T
p
M and g C

(N).
In the special case N = R, i.e. f : M R is a smooth function, we can identify
T
f(p)
R with R. Then we have
X
p
(f) = df
p
(X
p
).
In other words, df
p
T

p
M. We will call df
p
a cotangent vector at p.
We now set TM =
p
T
p
M, the disjoint union of all tangent vectors. It is called
the tangent bundle of M. There is a natural projection map
: TM M, (x, v) x.
Obviously we have T
x
M =
1
(x).
Proposition 2.9. TM is a smooth manifold of dimension 2n, and is smooth.
Proof. Suppose , U, V is a chart of M, then T,
1
U, V R
n
is a chart of
TM, where
T = ( , d) :
1
U V R
n
.
Similarly T

M =
p
T

p
M, the cotangent bundle of M, is also a smooth manifold
of dimension 2n, with the natural projection map a smooth map.
14
3 Vector Fields
3.1 Vector Fields
Denition 3.1. A vector eld is a section of the tangent bundle TM, i.e. a map
X : M TM such that X = Id
M
. It is smooth if for any f C

(M), the
function
Xf(p) = X
p
(f)
is a smooth function on M. The set of smooth vector elds on M is denoted by

(TM).
From now on when we say vector elds, we always mean smooth vector elds.
We can think of a vector eld X as a map
X : C

(M) C

(M), f Xf.
Consider two vector elds X and Y on M.
Lemma 3.2. At each point p M the bracket [X, Y ]
p
dened by
[X, Y ]
p
(f) = X
p
(Y f) Y
p
(Xf)
is a tangent vector at p.
Proof. We only need to check the Leibnitz law:
a[X, Y ]
p
(fg) =X
p
(Y (fg)) Y
p
(X(fg))
=X
p
((Y f)g +f(Y g)) Y
p
((Xf)g +f(Xg)
=X(Y f)(p)g(p) + Y f(p)Xg(p) +Xf(p)Y g(p) + f(p)X(Y g)(p)
Y (Xf)(p)g(p) Xf(p)Y g(p) Y f(p)Xg(p) f(p)Y (Xg)(p)
=f(p)[X, Y ]g(p) + [X, Y ]f(p)g(p).
Thus for any vector elds X and Y , the commutator [X, Y ] is still a vector eld.
More over, if X and Y are smooth, [X, Y ] is also smooth. It is called the Lie bracket
of vector elds X and Y . It is easy to see that the Lie bracket [X, Y ] satises the
following properties:
Proposition 3.3. (a)(skew symmetry)[X, Y ] = [Y, X].
(b)(R linearity)[aX +bY, Z] = a[X, Z] + b[Y, Z].
(c)(Jacobi identity)[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.
15
Proof. (a), (b) are obvious, and (c) follows from direct computation.
Denition 3.4. A Lie algebra g is a vector space together with a binary operation
[, ] : g g g, (v
1
, v
2
) [v
1
, v
2
]
which is bilinear, skew symmetric and satises the Jacobi identity.
Recall that for any function f : M N, f(p) = q, the dierential df
p
is a
push-forward map
df
p
: T
p
M T
q
N, (df
p
(X))g = X(g f).
A natural question is: Can we dene such a push-forward operator for vector
elds? There are two issues here. First if f is not surjective, then there is no
reasonable way to dene the push-forwarded vector eld on those points that are
not in the image of f. Second if f is not injective, then if f(p
1
) = f(p
2
), it is
possible that df
p
1
X(p
1
) ,= df
p
2
X(p
2
). However, if f : M N is a dieomorphism,
and X

(TM), one can push-forward X to a vector eld on N by

X((p)) = (d
p
)(X
p
).
A more general conception is
Denition 3.5. Suppose : M N is a smooth map, X is a vector eld on M
and Y is a vector eld on N. We say that X and Y are -related if for any p M
and q = (p), d
p
(X
p
) = Y
q
.
Lemma 3.6. Suppose : M N is smooth and X

(TM), Y

(TN) are
-related. then for or any f C

(N), X

f =

(Y f).
Proof. Suppose q = (p), then

(Y f)(p) = Y f(q) = Y
q
f = (d
p
X
p
)f = X
p
(f ) = X
p
(

f).
Corollary 3.7. If : M N is a dieomorphism,

Xf = (
1
)

f.
Proof. Obviously if : M N is a dieomorphism, X is -related to

X.
Corollary 3.8. If X
i
are -related to Y
i
for i = 1, 2, then [X
1
, X
2
] is -related to
[Y
1
, Y
2
].
Proof. For any g C

(N),
d
p
([X
1
, X
2
]
p
)(g) = X
1
(X
2
(

g))(p) X
2
(X
1
(

g))(p)
=

Y
1
(Y
2
(g))(p)

Y
2
(Y
1
(g))(p)
= Y
1
(Y
2
(g))((p)) Y
2
(Y
1
(g))((p))
= ([Y
1
, Y
2
]g)((p)).
16
3.2 Integral Curves
Recall that a curve in a smooth manifold M is a smooth map : I M, where
I is an interval in R. For any a I, the derivative of at t = a gives a tangent
vector on M at (a) by
d
dt
(a) = d
a
(
d
dt
).
Denition 3.9. We say that : I M is an integral curve of a vector eld X if
for any a I,
d
dt
(a) = X((a)).
Remark. Locally the equation
d
dt
(a) = X((a)) is a system of rst order dierential
equations. By Picard theorem, locally the integral curve always exists and unique,
and depends on the initial data smoothly.
Lemma 3.10. Suppose : M N is smooth and X

(TM), Y

(TN) are
-related. If is an integral curve of X, then is an integral curve of Y .
Proof. The chain rule implies
d
dt
(a) = d( )
a
(
d
dt
) = d
(a)
d
a
(
d
dt
) = d
(a)
X((a)) = Y ( (a)).
Denition 3.11. A vector eld X on M is complete if for any p M, there is an
integral curve : R M such that (0) = p.
As in the case of functions, we can dene the support of a vector eld by
supp(X) = p M [ X(p) ,= 0.
Theorem 3.12. If X is a compactly supported vector eld on M, then it is complete.
Proof. Let C = supp(X). Then any integral curve starting at p M C stays at
p. Thus every integral curve starting at p C stays in C. It follows that for any
p C, there is an interval I
p
0, a neighborhood U
p
of p in C and a smooth map
: U
p
I
p
C
such that for all q U
p
,

q
(t) = (q, t)
is an integral curve and
q
(0) = q. Since
p
U
p
= C, and C is compact, one can
extract a nite subcover U
p
1
, , U
pn
. Let I =
k
I
p
k
, then for any p C, there is
an integral curve
p
: I C. Since the interval I is xed, by uniqueness, one can
nd an integral curve curve : R C with (0) = p.
Corollary 3.13. Any smooth vector eld on a compact manifold is complete.
Proof. The set Supp(X), as a closed set in the compact manifold, is compact.
17
3.3 Dynamics of Vector Fields
Now suppose X is a complete vector eld on M. Then for any p M, there is
a unique integral curve
p
: R M such that
p
(0) = p. From this one can, for any
t R, dene a map

t
: M M, p
p
(t).
Notice that for any p M and any t, s R,
t

s
(p) and
t+s
(p) are both integral
curves for X with initial conditions
s
(p) at t = 0. By uniqueness, we have

t

s
=
t+s
.
Since
0
= Id, we conclude that
t
: M M is bijective.
Lemma 3.14. The map
: R M M, (t, p)
t
(p)
is smooth.
Proof. First by Picard theorem, the integral curves depends on the initial condition
smoothly. In other words, for any p M, there is a neighborhood U
0
of p and
an interval I
0
= (
0
,
0
) 0 such that [
I
0
U
0
is smooth. To show that is
smooth near any (p, t
0
), we notice that
t
(p) =
p
(t) is smooth on t, so the set

p
([
0
, t
0
+
0
]) is compact. The compactness argument allows us to take a smaller
interval I = (, ) such that is smooth in U I, where U is a neighborhood of

p
([
0
, t
0
+
0
]). It follows that if [t t
0
[ < ,
(t, p) = (t
0
+s, p) = (t
0
/N, (t
0
/N, (t
0
/N, (s, p))))
is smooth.
It follows that
t
are smooth maps. In other words, the family of maps
t
is
a family of dieomorphisms of M. Notice that
t

s
=
t+s
can be rewritten as
(t +s, p) = (t, (s, p)).
Denition 3.15. We will call : R M M, (t, p)
t
(p) the ow of X.
Lemma 3.16. Suppose X is complete, then for any f C

(M),
d
dt
[
t=0

t
f = Xf.
Proof. Let
p
(t) be the integral curve with
p
(0) = p, then

t
f(p) = f(
t
(p)) =
f(
p
(t)). Thus
d
dt
[
t=0

t
f(p) =
d
dt
[
t=0
f(
p
(t)) = df
p
(
d
dt
(
p
(t))[
t=0
) = df
p
(X
p
) = Xf(p).
18
Now we introduce the dynamical system description of Lie bracket [X, Y ]. Sup-
pose X is complete, and
t
is the family of dieomorphisms generated by X. Let
Y
t
= (
t
)

Y .
Theorem 3.17. [X, Y ] =
d
dt
[
t=0
Y
t
.
Proof. We need to show that for any f C

(M), [X, Y ]f =
d
dt
[
t=0
Y
t
f. Since
Y
t
= (
t
)

Y and each
t
is dieomorphism, we have Y
t
f = (
1
t
)

t
f. So
d
dt
[
t=0
Y
t
f(p) =
d
dt
[
t=0
(
1
t
)

t
f(p)
=
d
dt
(
t
)

Y f(p)[
t=0
+
d
dt
Y

t
f(p)[
t=0
= XY f(p) +Y Xf(p).
Suppose X is a complete vector eld on M. Let
t
be the family of dif-
feomorphisms generated by X. Notice that it is in fact a one-parameter group:

t

s
=
t+s
. We will denote
t
= exp(tX) to emphasis the X-dependence, and so
that the group law now reads
exp(tX) exp(sX) = exp((s +t)X).
Note that in general exp(tX) exp(sY ) ,= exp(sY ) exp(tX).
Theorem 3.18. Let X, Y be complete vector elds on M. Then exp(tX) commutes
with exp(sY ) for all t and s if and only if [X, Y ] = 0.
Proof. First from theorem 3.4 we see
(exp aX)

[X, Y ] =
d
dt
[
t=a
(exp tX)

Y
for all a. So
[X, Y ] = 0 (exp tX)

Y = Y, t
Y is exp tX related to Y itself
Both (s) = (exp tX)(exp sY )(p) and (s) = (exp sY )(exp tX)(p)
are integral curves of Y starts at (exp tX)(p)
(exp tX)(exp sY )(p) = (exp sY )(exp tX)(p).
19
4 Submanifolds
4.1 Submersions and Immersions
Let M, N be smooth manifolds and f : M N a smooth map, p M. Recall
that df : T
p
M T
f(p)
N is the linear map dened by
df(X
p
)(g) = X
p
(g f).
Denition 4.1. (1) f is a submersion at p if df
p
: T
p
M T
f(p)
N is surjective.
(2) f is an immersion at p if df
p
: T
p
M T
f(p)
N is injective.
Obviously if f is a submersion, then dimM dimN. Conversely, if f is an
immersion, then dimM N.
Theorem 4.2 (Canonical Submersion Theorem). Let f : M N be a submersion
at p M, then m = dimM n = dimN, and there exists charts (
1
, U
1
, V
1
)
around p and (
2
, U
2
, V
2
) around q = f(p) such that the map
2
f
1
1
: V
1
V
2
is exactly the projection map
: R
m
R
n
, (x
1
, , x
m
) (x
1
, , x
n
)
restricted to V
1
.
Theorem 4.3 (Canonical Immersion Theorem). Let f : M N be an immersion
at p M, then m = dimM n = dimN, and there exists charts (
1
, U
1
, V
1
)
around p and (
2
, U
2
, V
2
) around q = f(p) such that the map
2
f
1
1
: V
1
V
2
is exactly the inclusion map
: R
m
R
n
, (x
1
, , x
m
) (x
1
, , x
m
, 0, , 0)
restricted to V
1
.
Both of them are corollaries of
Theorem 4.4 (Inverse Mapping Theorem). Suppose M and N are both smooth
manifolds of dimension n, and f : M N a smooth map. Let p M, and
q = f(p) N. If df
p
: T
p
M T
q
N is bijective, then f maps a neighborhood U
1
of
p dieomorphically to a neighborhood U
2
of q.
Proof. Since the theorem is local, we can work on local charts, in which case the
theorem is a well known result in advanced calculus.
20
The idea that use local charts to translate theorems on R
n
to manifolds is very
useful. Similarly one can prove the chain rule: Let f : M
1
M
2
and g : M
2
M
3
be dierential maps, then for any p M
1
,
d(g f)
p
= dg
f(p)
df
p
.
Now we can prove the canonical theorems above.
Proof of the Canonical Submersion Theorem:
Since the problem is local, we might assume p = 0 U R
m
and q = 0
V R
n
. The fact that f is a submersion is equivalent to that the Jacobian matrix
(
f
i
x
j
) is an m n matrix of rank n. So in particular, m n. By reordering the
coordinates if necessary, we may assume the submatrix
(
f
i
x
j
), 1 i n, 1 j n
is nonsingular. Dene
F : U R
m
, (x
1
, , x
m
) (f
1
, , f
n
, x
n+1
, , x
m
).
Then obviously dF(0) is nonsingular. By inverse function theorem, locally F is
invertible with inverse H. Note that with the map dened in Theorem 1.2, f =
F. So
f H = F H = .
Proof of the Canonical Immersion Theorem:
Similarly we assume p = 0 U R
m
, q = 0 V R
n
, and the submatrix
(
f
i
x
j
), 1 i m, 1 j m
is nonsingular. Dene
F : UR
nm
R
n
, (x
1
, , x
m
, y
1
, , y
nm
)

f(x)+(0, , 0, y
1
, , y
nm
).
Then dF(0) is nonsingular. So locally F is invertible with inverse H, and
H f = H F = .
21
4.2 Submanifolds
Denition 4.5. Let M be a smooth manifold of dimension n. A subset S
M is a k dimensional (embedded/regular) submanifold of M if for every x S,
there is a chart (, U, V ) around x such that (U S) = V

R
k
, where

R
k
=
(x
1
, , x
k
, 0, , 0) R
n
.
Remark. If S is a k dimensional submanifold of M, then
S is a k dimensional manifold under the relative topology.
The inclusion map : S M is an immersion at all p S.
We may think of T
p
S as a vector subspace of T
p
M.
The denition is not convenient to use. In practice, the typical way to construct
submanifolds is to present them as images or level sets of smooth maps.
Denition 4.6. Let M, N be manifolds, and f : M N an immersion. f is called
an embedding if it is injective and proper.
Obviously if S is a submanifold of M, the inclusion : S M is an embedding.
Conversely,
Theorem 4.7. Suppose f : M N is an embedding, then f(M) is a submanifold
of N.
Proof. A well known fact from analysis is that any proper continuous map is a closed
map, i.e. it maps closed sets to closed sets. (Prove this if you dont know it.) So
if f : M N is an injective proper immersion, it is a homeomorphism from M to
f(M) N.
Now let p M and q = f(p). Since f is an immersion, the canonical immersion
theorem claims that there exists charts (, U, V ) and (
1
, U
1
, V
1
) such that on U,

1
f
1
is the canonical embedding : R
m
R
n
to the rst n coordinates. Since
f is a homeomorphism onto its image, f(U) is open in f(M). In other words, there
exists an open set

U
1
N such that f(U) = f(M)

U
1
. Replace U
1
by U
1

U
1
, and
V
1
by
1
(U
1

U
1
). Then
1
(U
1
f(M)) = V
1

1
(f(U)) = V
1
(U) = V
1

R
m
.
Denition 4.8. Suppose f : M N is a smooth map between smooth manifolds.
A point q N is called regular value if f is a submersion at each p f
1
(q).
Theorem 4.9. If q is a regular value of a smooth map f : M N, then S = f
1
(q)
is a submanifold of M of dimension dimM dimN. Moreover, for every p S,
T
p
S is the kernel of the map df
p
: T
p
M T
q
N.
22
Proof. Let p S = f
1
(q). Then by canonical submersion theorem, there are
charts (, U, V ) centered at p and (
1
, U
1
, V
1
) centered at q such that f(U) U
1
,
and =
1
f
1
. It follows that maps U f
1
(q) onto V
1
(0). So f
1
(q)
is a submanifold.
Now denote the inclusion by : S M. Then for any p S, f (p) = q. In
other words, f is a constant map on S. So df
p
d
p
= 0, i.e. df
p
= 0 on the image
of d
p
: T
p
S T
p
M. By dimension counting we conclude that T
p
S coincides with
the kernel of df
p
.
Example: Consider the map
f : R
n+1
R, (x
1
, , x
n+1
) x
2
1
+ +x
2
n+1
.
Then S
n
= f
1
(1). Since the Jacobian
Jf(x) = 2(x
1
, , x
n+1
) ,= 0
on S
n
, we conclude that S
n
is an n-dimensional submanifold of R
n+1
. Moreover, for
any a = (a
1
, , a
n+1
) S
n
,
T
a
S
n
= v R
n+1
[ v a = 0.
Remark. More generally one has the following Constant Rank Level Set Theorem:
Let M, N be smooth manifold, and F : M N be a smooth map with constant
rank k. Then each level set of F is a closed submanifold of codimension k in M. In
particular, the level sets of a submersion are all submanifolds.
Later on we will also need to consider the following more general notion of
submanifolds:
Denition 4.10. Let M be a smooth manifold. An k dimensional immersed sub-
manifold of M is a subset S M endowed with a structure of k dimensional
manifold such that the inclusion map : S M is an immersion.
Remark. (1) According to the canonical immersion theorem, the local structure of
an immersed submanifold is as good as a regular submanifold.
(2) Unlike submanifolds, the topology of the immersed submanifolds might be dier-
ent from the relative topology. For example, an immersed submanifold might have
self-intersecting.
(3) A very general constant rank theorem claims that, if f : M N is a smooth
map with constant rank (i.e. df is of constant rank at any point), then the image
of f is an immersed submanifold with tangent space the image of the tangent map.
23
4.3 Vector Fields on Submanifolds
Let S M be a submanifold. Since : S M is a smooth embedding, we
might think of T
p
S as a vector subspace of T
p
M for every p S. In other words,
any vector X
p
T
p
S can be identied with the vector

X
p
on M given by

X
p
f = (dX
p
)f = X(f ) = X
p
(f[
S
)
for any f C

(M).
Theorem 4.11. Suppose S M is a submanifold, and p S. Then
T
p
S = X
p
T
p
M [ X
p
f = 0 for all f C

(M) with f[
S
= 0.
Proof. Obviously if X
p
T
p
S, then for f C

(M) with f[
S
= 0, X
p
f = 0.
Conversely, if X
p
T
p
M satises X
p
f = 0 for all f that vanishes on S, we need
to show X
p
T
p
S. Take a coordinate chart (, U, V ) on M such that near p, S is
given by x
k+1
= = x
n
= 0. Then as we showed, T
p
M is the span of
1
, ,
n
.
Its obvious T
p
S is the subspace spanned by
1
, ,
k
. In other words, a vector
eld X
p
=

X
i

i
lies in T
p
S if and only if X
i
= 0 for i > k.
Now let h be a smooth bump function supported in U that equals 1 in a neigh-
borhood of p. For any j > k, consider the function f
j
(x) = h(x)x
j
, extended to be
zero on M U. Then f[
S
= 0. So
0 = X
p
f =

X
i
(h(x)x
j
)
x
i
= X
j
for any j > k. It follows that X
p
T
p
S.
Now suppose X is a vector eld on M. We say that X is tangent to a submanifold
S M if X
p
T
p
S T
p
M for every p S.
Corollary 4.12. A vector eld X

(TM) is tangent to a submanifold S M

if and only if Xf = 0 for every f C

(M) that vanishes on S.

Corollary 4.13. If vector elds X
1
, X
2

(TM) are tangent to a submanifold

S M, so is [X
1
, X
2
].
Remark. We can extend the conception of tangent to submanifold to immersed
submanifolds. Note that it is possible that all smooth vector elds are not tangent
to a given immersed submanifold. However, if Y

(M), and there is a vector

eld X

(S) that is -related to Y , then clearly Y is tangent to S. Conversely,

if Y

(M) is tangent to S, then there exists a vector eld on S that is -related

to Y . It follows that the above corollary holds for immersed submanifolds.
24
5 The Frobenius Theorem
5.1 Distributions
We have seen that any vector eld can be locally integrate to integral curves.
Note that if a vector eld is zero at some point, then the corresponding integral
curve is the constant path, i.e. whose image is a point. Similarly any nonzero vector
eld. as a rank 1 subbundle of the tangent bundle, will integrate to 1 dimensional
curve on the manifold. It is very nature to generalize these subbundle-submanifold
corresponding to higher dimensions.
Suppose M is an n-dimensional smooth manifold.
Denition 5.1. A k-dimensional distribution 1 on M is a map which assigns to
every p M a k-dimensional vector subspace 1
p
of T
p
M. 1 is called smooth if for
every p M, there is a neighborhood U of p and smooth vector elds X
1
, , X
k
on U such that for every q U, X
1
(q), , X
k
(q) are a basis of 1
q
.
For example, in R
n
, the vector elds

x
1
, ,

x
k
span a smooth k-dimensional
distribution. Note that they tangent to any k-dimensional planes that are parallel
to

R
k
.
Denition 5.2. Suppose 1 is a k-dimensional distributions on M. An immersed
submanifold N M is called an integral manifold for 1 if for every p N, the
image of d
N
: T
p
N T
p
M is 1
p
. 1 is integrable if through each point of M there
exists an integral manifold of 1.
Remarks. (1) Any nowhere vanishing smooth vector eld gives a smooth 1-dimensional
distribution. The image of any integral curve is an integral manifold.
(2) Unlike the 1 dimensional case that an integral curve always exists locally, an
integral manifold might not exists even locally. For example, consider the smooth
distribution on R
3
spanned by vector elds X =

x
+ y

z
and Z =

y
. Then there
is no integral manifold through the origin.
(3)An integral manifold doesnt have to be an embedded submanifold. For
example, consider M = S
1
S
1
R
2
R
2
. Fix any irrational number a, the
integral manifold of
X
a
= (x
2

x
1
x
1

x
2
) +a(y
2

y
1
y
1

y
2
) =

1
+a

2
is dense in M. (However, it is an immersed submanifold.)
We are interested in when a distribution has an integral manifold. Suppose N
is an integral manifold for 1, and X
p
, Y
p
1
p
, which can be identied with T
p
N.
25
Then we necessarily have [X
p
, Y
p
] T
p
N. In other words, an integrable distribution
1 must satises the following
Frobenius Condition: If X, Y

(M) belong to 1, so is [X, Y ].

Denition 5.3. A distribution 1 is involutive if it satises the Frobenius condition.
The arguments above proves
Proposition 5.4. Any integrable distribution is involutive.
Example: Any 1 dimensional distribution is involutive. The k-dimensional distri-
bution spanned by

x
1
, ,

x
k
is involutive. However, the distribution spanned by
X =

x
+y

z
and Y =

y
is not involutive, since [X, Y ] =

z
.
5.2 The Frobenius Theorem
Theorem 5.5 (Local Frobenius Theorem). Let 1 be an involutive m-dimensional
distribution. Then for every p M, there exists a coordinate patch (U, x
1
, , x
n
)
centered at p such that for all q U, 1
q
= span

x
1
(q), ,

xm
(q).
Before we proof the theorem, lets rst prove the special case for m = 1.
Lemma 5.6. Let X be a smooth vector eld on M. If p M such that X
p
,= 0,
then there exists a local chart (U, x
1
, , x
n
) near p such that X =

x
1
on U.
Proof. Choose a local chart (V, y
1
, , y
n
) around p such that X
p
=

y
1

p
. Denote
X =

y
i
on V , where
i
are smooth functions on V . Shrinking V if necessary,
we may assume
1
,= 0 on V . Consider the system of ODEs
dy
k
dy
1
=

k
(y
1
, , y
n
)

1
(y
1
, , y
n
)
, 2 k n.
By basic theory of ODE, locally for any given initial condition (z
2
, , z
n
), [z
k
[ < ,
the system above has a unique solution
y
k
=
k
(y
1
; z
2
, , z
n
), [y
1
[ <
with initial condition

k
(0; z
2
, , z
n
) = z
k
,
and the functions
k
depends smoothly on y
1
and on z
k
s. Make the change of
variables
y
1
= z
1
,
y
k
=
k
(z
1
; z
2
, , z
n
), 2 k n.
26
Since the Jacobian
(y
1
, , y
n
)
(z
1
, , z
n
)

z
1
=0
= 1.
there exists a neighborhood U V of p, with z
i
local coordinate functions. We have
in this chart
X =

y
i
=
1

y
i
z
1

y
i
=
1

z
1
.
Finally if we let x
1
=
_
z
1
0
dz
1

1
and x
k
= z
k
for k 2, then x
1
, , x
n
are local
coordinate functions on U such that X =

x
1
.
Proof of the Local Frobenius Theorem: We have proved this for m = 1.
Suppose the theorem holds for m 1 dimensional distributions. Let 1 be an m
dimensional distribution spanned by X
1
, X
2
, , X
m
. Suppose 1 is involutive, i.e.
[X
i
, X
j
] 0 mod X
k
, 1 i, j, k m.
Use the previous lemma, there exits a local chart near p such that X
m
=

xm
. For
1 k m1 let
X

k
= X
k
(X
k
x
m
)X
m
,
then X

k
x
m
= 0 for 1 k m 1, and X
m
x
m
= 1. Note that the vector elds
X

1
, , X

m1
, X
m
still span 1. Moreover, if we denote
[X

i
, X

j
] = a
ij
X
m
mod X

k
, 1 i, j, k m1,
then applying both sides to the function x
m
, we see a
ij
= 0 for all 1 i, j m1.
In other words, the m1 dimensional distribution
1

= spanX

1
, , X

m1

is involutive. So there is a local chart (U, z

1
, , z
n
) near p such that 1

is spanned
by

z
1
, ,

z
m1
. Since each

z
i
is a linear combination of X

j
for 1 i, j m1,
we conclude

z
i
x
m
= 0.
Now suppose
[

z
i
, X
m
] = b
i
X
m
mod

z
i
, 1 i m1.
Apply both sides to the function x
m
, we see b
i
= 0. So
[

z
i
, X
m
] =

C
m
i

z
i
.
27
Suppose X
m
=

z
j
. Insert this into the previous formula, we see

j
z
i
= 0, 1 i m1, m j n.
In other words, = (z
m
, , z
n
). Let X

m
=

n
k=m

z
k
, then

z
1
, ,

z
m1
, X

still span 1. Finally according to the pervious lemma again, there is a local coordi-
nate change from (z
m
, , z
n
) to (w
m
, , w
n
) such that X

m
=

wm
. This completes
the proof.
So any involutive distribution is integrable, i.e. locally near each point one can
nd an integrable manifold.
Theorem 5.7 (Global Frobenius Theorem). Let 1 be an involutive k-dimensional
distribution. Then through every point p M, there is a unique maximal connected
integral manifold of 1.
Sketch of proof: For any p M, let
N
p
= q M [ a piecewise smooth integral curve in 1 jointing p to q.
We claim that N
p
is the maximal connected integral manifold of 1 containing p.
The manifold structure is dened as follows: for any q N
p
, there is a coordinate
patch (U, x
1
, , x
n
) centering at q such that 1 = span

x
1
, ,

x
k
in U. For each
small , let
W = w U [ x
2
1
(w) + +x
2
k
(w) , x
k+1
(w) = = x
n
(w) = 0.
Then any point w W can be joint to p by the integral curve
(t) = t(x
1
(w), , x
k
(w), 0, , 0).
So W N
p
. Let
: W B
k
() R
k
, w (x
1
(w), , x
k
(w)).
Now we dene the topology on N
p
by giving it the weakest topology such that
the domains of all these maps are open. The atlas on N
p
is dened to be the set of
charts (, W, B
k
()). One can check that N
p
is a manifold with this given atlas. For
more details, c.f. Warner, pg.48-49.
Example: Consider the distribution on R
3
spanned by X = x

y
y

x
and Y =

z
on R
3
x = y = 0. Since [X, Y ] = 0, it is involutive. What is its integral
28
manifold? Well, lets rst compute the integral curves of X and Y . Through any
point (x, y, z), the integral curve of X is circle in the z-plane with origin the center,
and the integral curve of Y is the line that is parallel to the z-axis. Note that the
integral manifold passing (x, y, z) of the distribution should contains all points of
the form
X
t
(
Y
s
(x, y, z)) for all t, s. In our case, this is the cylinders centering with
the z-axis.
29
Part II
Part II: Basic Lie Theory
30
6 Lie Groups and their Lie Algebras
6.1 Lie Groups
Denition 6.1. A Lie group G is a smooth manifold equipped with a group struc-
ture so that the group multiplication
: GG G, (g
1
, g
2
) g
1
g
2
is a smooth map.
Here are some basic examples:
R
n
, considered as a group under addition.
R

= R 0, considered as a group under multiplication.

S
1
, Considered as a group under multiplication.
If M and N are Lie groups, so is their product M N.
Remarks. (1) The famous Hilberts fth problem, solved by Gleason and Montgomery-
Zippin in the 1950s, conrms that any locally Euclidean group (=topological group
whose underlying space is a topological manifold) is actually a Lie group, and
moreover, the underlying space of any Lie group is in fact an analytic manifold,
and the group operations are analytic. (An analytic manifold is a smooth manifold
whose transition maps are analytic. One can dene analytic maps between analytic
manifolds using local charts as in the smooth case.)
(2) Not every smooth manifold admits a Lie group structure. For example, the
only spheres that admit a Lie group structure are S
0
, S
1
and S
3
; among all the
compact 2 dimensional surfaces the only one admits a Lie group structure is T
2
=
S
1
S
1
. There are many constraints for a manifold to be a Lie group. For example,
the manifold must be analytic manifold. Two simple topological constraints are:
The tangent bundle of a Lie group is always trivial: TG G R
n
. In
particular, any Lie group is orientable.
The fundamental group of a connected Lie group is always Abelian.
Note that any Lie group is automatically a topological group. As in Lecture 1,
for any elements x, y G, there are two natural maps, the left multiplication
L
x
: G G, g x g
and the right multiplication
R
y
: G G, g g y.
It is obviously that both L
x
and R
y
are dieomorphisms, and they commutes with
each other: L
x
R
y
= R
y
L
x
.
31
Lemma 6.2. The dierential of the multiplication map : GG G is given by
d
x,y
(v, w) = (dR
y
)
x
(v) + (dL
x
)
y
(w), (v, w) T
x
GT
y
G.
Proof. Notice that as a function of x, (x, y) = R
y
(x), and as a function of y,
(x, y) = L
x
(y). Thus for any function f C

(G),
(d
x,y
(v, w))(f) = (v, w)(f (x, y))
= v(f R
y
(x)) + w(f L
x
(y))
= (dR
y
)
x
(v)(f) + (dL
x
)
y
(w)(f)
As an application, we can prove
Proposition 6.3. For any Lie group G, the group inversion map
i : G G, g g
1
is smooth.
Proof. Consider the map
f : GG GG, (x, y) (x, xy).
It is obviously a bijective smooth map. According to lemma above, the derivative
of f is
df
(x,y)
: T
x
GT
y
G T
x
GT
xy
G, (v, w) (v, (dR
y
)
x
(v) + (dL
x
)
y
(w))
This is a bijective linear map since dR
y
, dL
x
are. It follows by inverse function
theorem that f is locally a dieomorphism near each pair (x, y). However, since
f is globally bijective, it must be a globally dieomorphism. We conclude that its
inverse,
f
1
: GG GG, (x, z) (x, x
1
z)
is a dieomorphism. Thus the inversion i, as the composition
G Ge GG
f
1
GG

2
G
is smooth.
Denition 6.4. A Lie group homomorphism between two Lie groups is a smooth
map which is also a homomorphism of groups.
32
Since any Lie group is a topological group, we have from proposition 3.2 in
lecture 1
Proposition 6.5. If G is a connected Lie group, and U an open neighborhood of
e G, then G =

n=1
U
n
.
As a consequence, any Lie group homomorphism : H G is determined by
its behavior near e H, provided H is connected. We will show more later: in fact
is determined by d
e
: T
e
H T
e
G!
6.2 Lie Algebras Associated to Lie Groups
Suppose G is a Lie group. From the left translations L
a
one can, for any v T
e
G,
dene a vector eld X
v
on G by
X
v
(a) = (dL
a
)(v).
It is not surprising that X
v
is invariant under any left translation:
(dL
a
)(X
v
(b)) = dL
a
dL
b
(v) = dL
ab
(v) = X
v
(ab) = X
v
(L
a
b).
Denition 6.6. A left invariant vector eld on a Lie group G is a vector eld X
on G which satises (dL
a
)(X(b)) = X(L
a
b).
We will denote the set of left invariant vector elds on Lie group G by g, i.e.
g = X

(TG) [ X is left invariant.

This is a vector subspace of

(TG).
Proposition 6.7. For any X g, there is a v T
e
G such that X = X
v
.
Proof. Let v = X(e) T
e
G. Then for any a G, X(a) = (dL
a
)X(e) = (dL
a
)v.
So as a vector space, g is isomorphic to T
e
G.
Proposition 6.8. If X, Y g, so is their Lie bracket [X, Y ].
Proof. We have already seen that [X, Y ] is a vector eld on G. To show that it is
left-invariant, we rst notice
Y (f L
a
)(b) = (dL
a
Y
b
)f = Y (ab)f = (Y f)(L
a
b) = (Y f) L
a
(b)
for any smooth function f C

(G). Thus
X
ab
(Y f) = (dL
a
X
b
)Y f = X
b
((Y f) L
a
) = X
b
Y (f L
a
).
33
Similarly Y
ab
Xf = Y
b
X(f L
a
). Thus
dL
a
([X, Y ]
b
)f = X
b
Y (f L
a
) Y
b
X(f L
a
) = X
ab
(Y f) Y
ab
Xf = [X, Y ]
ab
(f).
It follows that the space g of all left invariant vector elds on G together with the
Lie bracket operation [, ] is a Lie subalgebra of the Lie algebra of all smooth vector
elds

(TG). Notice that as a vector space,

(TG) is innitely dimensional,

while g T
e
G is n-dimensional.
Denition 6.9. g of is called the Lie algebra of G.
Now suppose : G H is a Lie group homomorphism, then its dierential at
e gives a linear map from T
e
G to T
e
H. Under the identication of T
e
G with g and
T
e
H with h, we get an induced map, still denoted by d, from g to h.
Theorem 6.10. If : G H is a Lie group homomorphism, then the induced map
d : g h is a Lie algebra homomorphism.
Proof. We need to show that d preserves the Lie bracket. Let X
v
be a left in-
variant vector eld on G, then by denition d(X
v
) = X
de(v)
. Since is a group
homomorphism, we have L
g
= L
h
. It follows
(d
g
)(X
v
)
g
= (d
g
)(dL
g
)
e
v = d(L
g
)
e
v = d(L
h
)
e
v = dL
h
(d
e
)v = X
de(v)
(h) = d(X
v
)(h).
In other words, X
v
and d(X
v
) are -related. It follows that [X
v
, X
w
] and [d(X
v
), d(X
w
)]
are -related. On the other hand, [X
v
, X
w
] is -related to d([X
v
, X
w
]). So
[d(X
v
), d(X
w
)]
e
= d
e
([X
v
, X
w
]
e
) = (d([X
v
, X
w
]))
e
,
i.e. d([X
v
, X
w
]) = [d(X
v
), d(X
w
)].
6.3 Examples
Example 1: The Euclidean group R
n
.
This is obviously a Lie group, since the group operation
((x
1
, , x
n
), (y
1
, , y
n
)) := (x
1
+y
1
, , x
n
+y
n
).
is smooth.
Moreover, for any a R
n
, the left translation L
a
is just the normal translation
map on R
n
. So dL
a
is the identity map, as long as we identify T
x
R
n
with R
n
in the
34
usual way. It follows that any left invariant vector eld is in fact a constant vector
eld, i.e.
X
v
= v
1

x
1
+ +v
n

x
n
for v = (v
1
, , v
n
) T
0
R
n
. Since

x
i
commutes with

x
j
for any pair (i, j), we
conclude that the Lie bracket of any two left invariant vector elds vanishes. In
other words, the Lie algebra of R
n
is R
n
with vanishing Lie bracket.
Example 2: The circle S
1
The Lie group multiplication
(e
it
1
, e
it
2
) = e
i(t
1
+t
2
)
is obviously smooth. Moreover, the left translation maps are just rotation maps. The
derivative of these maps are also the identity map. It follows that the Lie algebra
of S
1
is R
1
with vanishing Lie bracket. (In fact, from denition of Lie algebra, it is
obvious that the only 1-dimensional Lie algebra is this trivial Lie algebra.)
Example 3: The ane group R

R
1
One can equip with R

R the usual manifold structure and with the group

operation
((x
1
, y
1
), (x
2
, y
2
)) = (x
1
x
2
, x
1
y
2
+y
1
).
This is called the ane group of R
1
. It is obviously a non-abelian Lie group.
Taking derivative from the previous expression of L
(x
1
,y
1
)
(x
2
, y
2
), we get
dL
(x
1
,y
1
)
(v
1

x
+v
2

y
) = x
1
v
1

x
+x
1
v
2

y
.
Thus if a(x, y)

x
+b(x, y)

y
is a left invariant vector eld, we must have
a(x
1
x, x
1
y +y
1
) = a(x, y)x
1
and b(x
1
x, x
1
y +y
1
) = b(x, y)x
1
for all x, y, x
1
, y
1
R. In particular, X = x

x
and Y = x

y
are two left invariant
vector elds. Since the Lie group is 2-dimensional, X, Y form a basis of its Lie
algebra. The Lie brackets are given by [X, X] = [Y, Y ] = 0 and
[X, Y ] = [x

x
, x

y
] = x

y
= Y.
35
7 The exponential map
7.1 One Parameter Subgroups
Recall that R, as an additive group, is a Lie group.
Denition 7.1. A one parameter subgroup of a Lie group G is a Lie group homo-
morphism : R G, i.e. is smooth such that (s +t) = (s)(t).
Let : R G be a one parameter subgroup of G. Using right multiplication
one can dene
: R G G, (t, g) g (t).
This is a ow on G, i.e.
(t +s, g) = g (t +s) = g ((t)(s)) = (s, (t, g)).
Since the right multiplication commutes with left multiplication, the ow dened
above is left invariant,
L
a
((t, g)) = a(g (t)) = (a g)(t) = (t, L
a
g).
Such ows on G are called left invariant ows.
Theorem 7.2. There are one-to-one correspondences between
1. One parameter subgroups of G.
2. Left invariant ows on G.
3. Left invariant vector elds on G.
4. Tangent vectors at e G.
Proof. We have seen (3)(4) in lecture 6.
(1)(2): For any one parameter subgroup of G, we just constructed a left
invariant ow on G. Conversely, if is a left invariant ow on G, we dene : R G
by (t) = (t, e). It follows that (t + s) = (t)(s), i.e. : R G is a one
parameter subgroup. Obviously, the two procedure and are inverse
to each other.
(1)(4): Suppose : R G is a one parameter subgroup of G. Then we can
dene a vector at e G by v = (d)
0
(
d
dt
). Conversely, for any vector v T
e
G,
one can generate a unique left invariant vector eld X
v
on G such that X
v
(e) = v.
Since any left invariant vector eld is complete, for any g G, one can nd an
36
integral curve
g
: R G of X
v
such that
g
(0) = g. We thus get a map :
R G G, (t, g) =
g
(t). We have seen in lecture 3 that is a ow. It is also
left invariant since X
v
is left invariant:
d(L
a

g
(t))(
d
dt
) = (dL
a
)
d
g
dt
= (dL
a
)X
v
(g) = X
v
(ag)
implies that L
a

g
(t) is the integral curve of X
v
starts at ag, i.e. a
g
(t) =
ag
(t).
Obviously
v X
v
and v X
v
v
are both identity maps.
7.2 The Exponential Map
Denition 7.3. The exponential map of G is the map
exp : g G, X
X
(1),
where
X
is the 1-parameter subgroup of G corresponding to X.
Obviously

(s) =
X
(ts) is the one parameter subgroup corresponding to tX,
we have
exp(tX) =
X
(t).
Example: (1) For G = R

, X T
1
G = R, exp(tX) = e
tX
.
(2) For G = T = S
1
, g = R. exp is the map exp(tX) = e
itX
.
(3) For G = R, X T
0
G = R, exp(tX) = tX.
(4) For G = GL(n), X T
e
G = M(n), exp(tX) = e
tX
.
Lemma 7.4. exp : g G is a smooth map, and if we identify T
0
g with g, the
tangent map of exp at 0 is the identity map.
Proof. Consider the map

: R Gg Gg, (t, g, X) (g exp(tX), X).

One can check that this is the ow on G g corresponding to the left invariant
vector eld (g, X) (X(g), 0) on G g, thus it is smooth. It follows that exp =

[
{1}{e}g
is smooth.
Since exp(tX) =
X
(t),
d
dt
[
t=0
exp(tX) = X. On the other hand,
d
dt

t=0
exp tX = (d exp)
0
d(Xt)
dt
= (d exp)
0
X.
We conclude that (d exp)
0
equals to the identity map.
37
Since (d exp)
0
is bijective, we have
Corollary 7.5. exp is a local dieomorphism near 0, i.e. it is a dieomorphism
from a neighborhood of 0 T
e
G to a neighborhood of e G.
Recall that for any Lie group homomorphism : G H, its dierential at e,
d : g h is a Lie algebra homomorphism.
Proposition 7.6 (exp is Natural). Given any Lie group homomorphism : G H,
the diagram
g
de
h

_
exp

_
exp
G

H
is commutative, i.e. exp
g
= exp
h
(d)
e
.
Proof. Let X g, then
exp
g
: R H, t exp
g
(tX)
is a one parameter subgroup of H, with
d
dt

t=0
exp
g
(tX) = d
e
(X).
So exp
g
(tX) = exp
h
t(d)
e
(X).
As an application, we have
Corollary 7.7. If G is connected, any Lie group homomorphism : G H is
determined by the induced Lie algebra homomorphism d : g h.
Proof. In a neighborhood of 0, exp is dieomorphism. Thus by the commutative
diagram, is determined by d in a small neighborhood of e. However, any such
neighborhood will generate the Lie group as long as the Lie group is connected. This
completes the proof.
7.3 Dierent Descriptions of the Lie Bracket
So we have four dierent set-theoretic descriptions of the Lie algebra g of G.
Consequently we should also have four dierent descriptions of the Lie bracket op-
eration [, ]:
38
(a) g = the set of left invariant vector elds on G: For left invariant vector elds
X and Y ,
[X, Y ] := XY Y X.
(b) g = T
e
G: For X, Y T
e
G,
[X, Y ] := ad(X)Y,
where ad : T
e
G End(T
e
G) is dened as follows. Each element g G gives
rise to an automorphism
c(g) : G G, x gxg
1
.
Notice that c(g) maps e to e, its dierential at e gives us a map
Ad
g
: T
e
G T
e
G.
In other words, we get the following adjoint representation
Ad : G End(T
e
G), g Ad
g
.
Note that Ad(e) is the identity map in End(T
e
G). Moreover, since End(T
e
G) is
a linear space, its tangent space at Id can be identied with End(T
e
G) itself in
a natural way. Taking derivative again at e, we get the adjoint representation
of Lie algebra
ad : T
e
G End(T
e
G).
Applying Proposition 2.4 to the Lie group homomorphism Ad : G End(g)
and to the conjugation map c(g) : G G, we have
Proposition 7.8. (1) Ad(exp(tX)) = exp(tad(X)).
(2) g(exp tX)g
1
= exp(tAd
g
X).
(c) g = the set of left invariant ows: Suppose X, Y are left invariant ows on
G. Let X
t
= X(t, ) : G G be the family of dieomorphism corresponding
to X, and still denote by Y the left invariant vector eld corresponding to the
left invariant ow Y . Set Y
t
= (X
t
)

Y . Then
[X, Y ] :=
d
dt
[
t=0
Y
t
.
We have seen in lecture 3 that this coincides with the denition in (a).
(d) g = the set of one parameter subgroups: The one parameter groups generated
by X, Y T
e
Gare exp(tX) and exp(tY ). Dene a(s, t) = exp(sX) exp(tY ) exp(sX).
Then
[exp(tX), exp(tY )] := exp(t

s

s=0

t=0
a(s, t)),
39
Now we show that the four dierent Lie bracket described at the end of section
1 are equivalent:
Theorem 7.9. The four dierent Lie brackets dened in (a), (b), (c), (d) of previous
section are equivalent.
Proof. We have seen that (a) is equivalent to (c). Now lets compute (adX)Y . The
previous arguments shows that
(adX)Y = (
d
dt

t=0
Ad(exp tX))Y =
d
dt

t=0
(Ad(exp tX)Y ).
On the other hand, since Ad
g
is the dierential of c(g), we have
Ad(exp tX)Y =
d
ds

s=0
c(exp tX) exp sY =
d
ds

s=0
exp(tX) exp(sY ) exp(tX).
This shows that (b) is equivalent to (d). To show that they are equivalent to (a),
we compute for any f C

(G),
(ad(X)Y )f = (
d
dt

t=0
(Ad(exp tX)Y ))f
=

2
st

s=t=0
f(exp(tX) exp(sY ) exp(tX))
=

2
st

s=t=0
f(exp(tX) exp(sY )) +

2
st

s=t=0
f(exp(sY ) exp(tX))
= XY f(e) Y Xf(e).
40
8 Linear Lie Groups
8.1 The General Linear Group
Let K = R or C. Recall that M(n, K), the set of n n K-valued matrices, is
dieomorphic to K
n
2
.
Denition 8.1. A Linear Lie group, or matrix Lie group, is a submanifold of
M(n, K) which is also a Lie group, with group multiplication the matrix multi-
plication.
Lets begin with the largest linear Lie group, the general linear group
GL(n, R) = X M(n, R) [ det X ,= 0.
Since GL(n, R) is open in M(n, R), and is closed under the group multiplication and
inversion operations, it is a Lie group. Obviously GL(n, R) is n
2
-dimensional and is
noncompact. Moreover, it is not connected, but consists of exactly two connected
components,
GL
+
(n, R) = A M(n, R) [ det A > 0 and GL

(n, R) = A M(n, R) [ det A < 0.

The fact that GL(n, R) is an open subset of M(n, R) R
n
2
also implies that the
Lie algebra of GL(n, R), as the tangent space at e = I
n
, is the set M(n, R) itself,
i.e.
gl(n, R) = A [ A is an n n matrix.
Given any A gl(n, R), we can dene
e
A
= I
n
+A +
A
2
2!
+
A
3
3!
+ +
A
n
n!
+ .
It is easy to check that the series converges, and
e
sA
e
tA
= e
(s+t)A
.
Notice that e
0A
= I
n
, we have (e
tA
)
1
= e
tA
. In particular, e
tA
GL(n, R). In
other words, e
tA
is a one parameter subgroup of GL(n, R). Since
d
dt
[
t=0
e
tA
= A, we
conclude that the exponential map exp : gl(n, R) GL(n, R) is
exp(A) = e
A
= I
n
+A +
A
2
2!
+
A
3
3!
+ .
As a corollary, we can compute the Lie bracket on gl(n, R) as
[A, B] =
d
dt

t=0
d
ds

s=0
(e
tA
e
sB
e
tA
) =
d
dt

t=0
(e
tA
Be
tA
) = AB BA.
41
Similarly if K = C, one can show that GL(n, C), the set of nonsingular n n
complex matrices, is a (connected) Lie group whose Lie algebra is gl(n, C), the set
of all n n complex matrices. Moreover, the Lie bracket on gl(n, C) is given by the
commutator
[A, B] = AB BA
and the exponential map is also the same expression as above.
8.2 The Special Linear Groups
The special linear group are dened as
SL(n, K) = X GL(n, K) : det X = 1.
Without loss of generality, we restrict ourselves to the case K = R. It is easy see
that SL(n, R) is a subgroup of GL(n, R). To show that it is a Lie subgroup, we
need to prove that SL(n, R) is a manifold. According to theorem 2.5 in lecture 4,
it suces to show that 1 is a regular value of the function det : GL(n, R) R,
i.e. d det is surjective at every point. Since SL(n, R) is a group, it suces to show
that d det is surjective at the identity I
n
. This is true, since by chain rule, for any
A gl(n, R),
(d det)
In
(A) =
d
dt

t=0
det(I
n
+tA) =
d
dt

t=0
(1 +tTrA +O(t
2
)) = TrA.
We conclude that SL(n, R) is a n
2
1 dimensional connected non-compact Lie group.
To determine its Lie algebra sl(n, R), we notice that as a vector space sl(n, R)
is a subspace of gl(s, R). Moreover, an n n matrix A lies in sl(n, R) if and only if
e
A
SL(n, R). On the other hand, basic linear algebra implies
det e
A
= e
TrA
.
It follows that e
A
SL(n, R) if and only if Tr(A) = 0. We conclude
sl(n, R) = A M(n, R) [ TrA = 0.
Everything is the same if we replace R by C.
8.3 The Orthogonal Groups and Unitary Groups
Next lets consider the orthogonal groups
O(n) = X GL(n, R) : X
T
X = I
n
.
42
This is another subgroup of GL(n, R). We claim that O(n) is an
n(n1)
2
dimensional
submanifold of M(n, R). To show this, we consider the map
f : M(n, R) S(n), A A
T
A,
where
S(n) = B M(n, R) [ B
T
= B
is the set of n n symmetric matrices. We want to show that I
n
is a regular value
of f. Notice that S(n) is an
n(n+1)
2
dimensional vector subspace of M(n, R). Under
the canonical identication, one can show that the map
df
A
: T
A
M(n, R) = M(n, R) T
f(A)S(n)
= S(n)
is given by
df
A
(C) = A
T
C +C
T
A.
Now suppose A f
1
(I
n
). Then for any B S(n), we have
df
A
(
AB
2
) = A
T
AB
2
+
B
T
A
T
2
A = B.
In other words, df
A
is surjective. We conclude that O(n) is an
n(n1)
2
dimensional
Lie subgroup of GL(n, R).
To gure out its Lie algebra o(n), we note that (e
A
)
T
= e
A
T
, so
(e
tA
)
T
e
tA
= I
n
e
tA
T
= e
tA
For all t. Since the exponential map is locally bijective, we conclude that A o(n)
if and only if A
T
= A. So
o(n) = A M(n, R) [ A
T
+A = 0,
the space of n n skew-symmetric matrices.
Notice that O(n) is not connected. It consists of two connected components,
and the connected component of identity is the called the special orthogonal groups
SO(n) = X GL(n, R) : X
T
X = I
n
, det X = 1 = O(n) SL(n, R).
Its Lie algebra so(n) is the same as o(n).
Similarly, the unitary groups
U(n) = X GL(n, C) :

X
T
X = I
n

43
is a Lie subgroup of GL(n, C) with Lie algebra
u(n) = A M(n, C) [

A
T
+A = 0,
the space of skew-Hermitian matrices.
Also the special unitary groups
SU(n) = U(n) SL(n, C)
has Lie algebra
su(n) = A M(n, C) [

A
T
+A = 0, Tr(A) = 0.
8.4 Symplectic Groups and Other Linear Lie Groups
Let J =
_
0 I
n
I
n
0
_
. The symplectic groups are by denition
Sp(2n, R) = X GL(2n, R) : X
T
JX = J.
It is a Lie group of dimension n(2n + 1) whose Lie algebra is
sp(2n, R) = A M(2n, R) [ JA +A
T
J = 0
= A =
_
A
1
A
2
A
3
A
4
_
[ A
i
M(n, R), A
1
= A
T
4
, A
2
= A
T
2
, A
3
= A
T
3
.
This, as well as orthogonal groups in the previous section, are special cases of the
following more general result. Let : R
n
R
n
R be a bilinear form on R
n
.
Consider the set of all invertible n n matrices that preserves ,
GL

(n, R) = X GL(n, R) [ (Xu, Xv) = (u, v)for all u, v R

n
.
In matrix form, there is a matrix B such that (u, v) = u
T
Bv. Then
GL

(n, R) = X GL(n, R) [ X
T
BX = B.
Lemma 8.2. GL

(n, R) is a linear Lie group with Lie algebra

gl

(n, R) = A M(n, R) [ A
T
B +BA = 0.
Proof. One can easily check that GL

(n, R) is a subgroup of GL(n, R), and it is

topologically a closed subset. By Cartans theorem we will prove later, it is a Lie
subgroup.
To describe its Lie algebra, notice that A gl

(n, R) if and only if e

tA

GL

(n, R), i.e. e

tA
T
Be
tA
= B. By taking t derivative at t = 0, we get A
T
B+BA = 0.
Conversely, if A
T
B + BA = 0, i.e. tA
T
B = B(tA), one can easily derive by
denition that e
tA
T
B = Be
tA
, i.e. e
tA
T
Ne
tA
= B. This completes the proof.
44
Notice that in the case B = I
n
( = the standard inner product), we get the
orthogonal groups O(n), and in the case B = J( = the standard symplectic form),
we get the symplectic groups above. If we take to be the standard inner product
of signature (p, n p),
(x, y) =
p

i=1
x
i
y
i

i=p+1
x
i
y
i
,
then we will get the indenite orthogonal group O(p, n p),
O(p, n p) = X GL(n, R) [ X
T
I(p, n p)X = I(p, n p),
where I(p, n p) = diag(I
p
, I
np
). Its Lie algebra is
o(p, n p) = X M(n, R) [ A
T
I(p, n p) = I(p, n p)A.
All the previous arguments holds if we replace R with C. In the later case, there is
a notion of compact symplectic group
Sp(n) = U(2n) Sp(2n, C).
This is a real compact Lie group of dimension n(2n + 1).
There are also many other interesting linear Lie groups, for example
T
n
: The set of diagonal nn complex matrices with unitary diagonal entries.
A
n
: The set of diagonal n n matrices with positive diagonal entries.
N
n
: The set of upper triangle nn matrices with all diagonal entries equal 1.
Heisenberg group H = A =
_
_
1 a b
0 1 c
0 0 1
_
_
[ a, b, c R.
Projective general linear group PGL(n, K) = GL(n, K)/Z(GL(n, K)).
Projective special linear group PSL(n, K) = SL(n, K)/Z(SL(n, K)).
Any intersection of above
We end this section by the following remark:
Remark. We will prove later that any compact Lie group G is a closed subgroup of
O(n) for n large enough. In particular, any compact Lie group is a linear Lie group.
(However, this property does not hold for noncompact Lie group. A well known
example is the metaplectic group Mp
2n
(the double cover of the symplectic group
Sp(2n, R)).)
45
9 The Baker-Campbell-Hausdor Formula
9.1 Taylor Series Expansion on Lie Groups
Let X g be a left invariant vector eld on G. Then
(Xf)(g) = X(g)f = dL
g
X(e)f = X
e
(f L
g
) =
d
dt

t=0
f(g exp(tX)).
for any f C

(G) and any g G. More generally,

Lemma 9.1. (1) Suppose X g. Then for any k 0 and any t R,
(X
k
f)(g) =
d
k
dt
k

t=0
f(g exp(tX)).
(2) If X
1
, , X
k
g ,then
(X
k
X
1
f)(g) =

k
t
1
t
k

t
1
==t
k
=0
f(g exp(t
1
X
1
) exp(t
k
X
k
)).
Proof. (1) This is the special case of
(X
k
f)(g exp(tX)) =
d
k
dt
k
f(g exp(tX))
which follows from induction and
(Xf)(g exp(tX)) =
d
ds

s=0
f(g exp(tX) exp(sX)) =
d
ds

s=0
f(g exp((t+s)X)) =
d
dt
f(g exp(tX)).
(2) Note that
(X
k
f)(g exp(t
1
X
1
) exp(t
k1
X
k1
)) =

t
k

t
k
=0
f(g exp(t
1
X
1
) exp(t
k1
X
k1
) exp(t
k
X
k
)).
The formula we want to prove follows from induction.
As a corollary, we get
Corollary 9.2. If f is a smooth function on G, then for small t,
f(exp(tX)) =
n

k=0
t
k
k!
X
k
f(e) +O(t
n+1
).
46
Note that the previous formulae hold for vector-valued functions as well. Our
main result in this section is
Theorem 9.3. Let n 1 and X
1
, , X
n
g. Then for [t[ suciently small,
exp(tX
1
) exp(tX
n
) = exp(t

1in
X
i
+
t
2
2

1i<jn
[X
i
, X
j
] +O(t
3
)).
Proof. Suppose f is smooth near e G. Then by lemma 1.1, for small t,
f(exp(tX
1
) exp(tX
n
)) = f(e)+t

i
X
i
f(e)+
t
2
2

i
X
2
i
f(e)+2

i<j
X
i
X
j
f(e)+O(t
3
).
We will apply this to the inverse of the exponential map near e, i.e. the map dened
by
f(exp(tX)) = tX
for t small enough. Then obviously, f(e) = 0. For any X g,
(Xf)(e) =
d
dt

t=0
f(exp(tX)) =
d
dt

t=0
(tX) = X
and for any n > 1,
(X
n
f)(e) =
d
n
dt
n

t=0
f(exp(tX)) =
d
n
dt
n

t=0
(tX) = 0.
Notice

i
X
2
i
+ 2

i<j
X
i
X
j
= (X
1
+ +X
n
)
2
+

i<j
[X
i
, X
j
],
it follows that
f(exp(tX
1
) exp(tX
n
)) = t

i
X
i
+
t
2
2

i<j
[X
i
, X
j
] + O(t
3
).
On the other hand, by the denition of f,
exp(tX
1
) exp(tX
n
) = exp(f(exp(tX
1
) exp(tX
n
))).
This completes the proof.
Corollary 9.4. For X, Y g and small t, we have
(1) exp(tX) exp(tY ) = exp(t(X +Y ) +
1
2
t
2
[X, Y ] +O(t
3
)).
(2) exp(tX) exp(tY ) exp(tX) = exp(tY +t
2
[X, Y ] + O(t
3
)).
(3) exp(tX) exp(tY ) exp(tX) exp(tY ) = exp(t
2
[X, Y ] +O(t
3
)).
Remark. The relation (3) gives a geometric interpretation of the Lie bracket: [X, Y ]
is the tangent vector at e to the curve t exp(

tX) exp(

tY ) exp(

tX) exp(

tY ).
47
9.2 The Derivative of the Exponential Map
We have seen that exp(sX) commutes with exp(tY ) for all t, s if and only if
[X, Y ] = 0. In other words, the Lie bracket measures the non-commutativity of
the Lie group multiplication. The results in the previous section is a quantitative
version of this statement: up to the second order, the Lie bracket measures the
non-commutativity of the Lie group multiplication. A natural question is: what are
the O(t
3
) terms? (Will they give new operations in Lie algebra?)
The answer to this problem is the Baker-Campbell-Hausdor Formula. In order
to prove the BCH formula, we need to compute the derivative of the exponential
map at an arbitrary point rst. (Recall we have seen that d exp
0
is the identity
map.)
The results in this section as well as in the next section holds for arbitrary Lie
groups. However, for simplicity we will assume that G is a connected Linear Lie
group.
To begin with, consider the function
(z) =
1 e
z
z
=

m=0
(1)
m
(m + 1)!
z
m
.
Lemma 9.5. For A, B M(n, R) and t R,
d
dt

t=0
e
A+tB
= e
A
(adA)B.
Proof. We have
e
A
d
dt

t=0
e
A+tB
=

r0
(1)
r
r!
A
r

n0
d
dt

t=0
(A +tB)
n
n!
=

r0,n0,0sn
(1)
r
r!(n + 1)!
A
r
A
s
BA
ns
=

r0,mr,rkm
(1)
r
(m + 1)!
_
m + 1
r
_
A
k
BA
mk
=

m0,0km,0rk
(1)
r
(m + 1)!
_
m + 1
r
_
A
k
BA
mk
=

m0,0km
(1)
k
(m + 1)!
_
m
k
_
A
k
BA
mk
,
where the third equality follows from the change of variables n m = r + n,
48
s k = r +s, the last equality follows from the combinatorial identity

0rk
(1)
r
_
m + 1
r
_
= (1)
k
_
m
k
_
.
On the other hand, since L
A
commutes with R
A
, we have
(adA)
m
B = (L
A
R
A
)
m
B =

0km
(1)
k
_
m
k
_
(L
A
)
k
(R
A
)
mk
B =

0km
(1)
k
_
m
k
_
A
k
BA
mk
.
The lemma follows.
Theorem 9.6. For each A g, (d exp)
A
= (dL
exp A
)
I
(adA).
Proof. For any B g,
(d exp)
A
(B) = (d exp)
A
d(A +tB)
dt
=
d
dt

t=0
e
A+tB
= e
A
(adA)B,
where e
A
, as a map from g = T
I
GL
n
to T
exp A
GL
n
, is exactly (dL
exp A
)
I
.
Corollary 9.7. The singular points of the exponential map exp : g G are precisely
the A g such that adA has an eigenvalue of the form 2ik with k Z 0.
9.3 The Baker-Campbell-Hausdor Formula
Denote by log the inverse of exp, at least for t small. Let
(t, A, B) = log(e
tA
e
tB
).
We have seen (t, A, B) = t(A +B) +
1
2
t
2
[A, B] + O(t
3
). In fact,
(t, A, B) = log(I +

p,q0,p+q>0
(tA)
p
(tB)
q
p!q!
)
=

k=1
(1)
k1
k

p
1
,q
1
, ,p
k
,q
k
0,p
i
+q
i
>0
(tA)
p
1
(tB)
q
1
(tA)
p
k
(tB)
q
k
p
1
!q
1
! p
k
!q
k
!
= t(A +B) +

m=2
t
m
P
m
(A, B).
For small m, we have
P
2
(A, B) =
1
2
[A, B], P
3
(A, B) =
1
12
([A, [A, B]][B, [A, B]]), P
4
=
1
24
[A, [B, [B, A]]].
We can see that each P
m
is a Lie polynomial, i.e. a combination of nested commu-
tators in A, B. It turns out that this is true in general:
49
Theorem 9.8 (The Baker-Campbell-Hausdor formula). P
m
(A, B) is a Lie poly-
nomial.
A very explicit formula is given by Dynkin:
Theorem 9.9 (Dynkins formula). For A and B small,
(A, B) = A+B+

k=1
(1)
k
k+1

(1)

i
(l
i
+m
i
)
l
1
+ +l
k
+1
(adB)
l
1
l
1
!

(adA)
m
1
m
1
!

(adB)
l
k
l
k
!

(adA)
m
k
m
k
!
(B),
where the second summation is over l
1
, , l
k
, m
1
, , m
k
0, l
i
+m
i
> 0.
Proof. Let C(t) = log(e
A
e
tB
), then
d
dt
e
C(t)
= e
C(t)
B. On the other hand, by lemma
2.1,
d
dt
e
C(t)
= e
C(t)
(adC(t))
dC
dt
.
It follows
dC
dt
=
adC(t)
I e
adC(t)
(B) =

k0
1
k + 1
(I e
adC(t)
)
k
(B).
Notice that for matrices, Ad(X)A = XAX
1
, so Ad(XY ) = Ad(X) Ad(Y ). Thus
dC
dt
=

k0
(I Ade
C(t)
)
k
k + 1
(B) =

k0
(I Ade
tB
Ade
A
)
k
k + 1
(B) =

k0
(I e
tadB
e
adA
)
k
k + 1
(B).
It follows
dC
dt
=

k0
(1)
k
k + 1

l
1
, ,l
k
,m
1
, ,m
k
0,l
i
+m
i
>0
t
|l|
(1)
|l|+|m|
(adB)
l
1
l
1
!
(adA)
m
1
m
1
!

(adB)
l
k
l
k
!
(adA)
m
k
m
k
!
B.
Now the formula follows from termwise integration over t from 0 to 1.
50
10 Lie Subgroups
10.1 Lie Subgroups v.s. Lie Subalgebras
Lemma 10.1. Let G be a Lie group, H G be both a subgroup and a smooth
submanifold. Then H is a Lie group.
Proof. Since : GG G is smooth, and HH GG is a smooth submanifold,
the map
H
: H H G is smooth. Since H is a subgroup,
H
maps into the
smooth submanifold H, hence is smooth as a map H H H.
Denition 10.2. A subgroup H of a Lie group G is called a Lie subgroup if it is
a Lie group (with respect to the induced group operation), and the inclusion map

H
: H G is a smooth immersion.
Note that we dont require H to be a smooth submanifold of G.
Example: Consider G = T
2
= S
1
S
1
. Then S
1
0 and 0 S
1
are Lie
subgroups. Moreover, for any co-prime pair of integers (p, q),
H
p,q
:= (e
ipt
, e
iqt
) [ t R
is a Lie subgroup of T
2
. These are submanifolds as well. However, there are also
many Lie subgroups of T
2
which are not submanifolds. In fact, for any irrational
number ,
H

:= (e
it
, e
it
) [ t R
is a Lie subgroup of T
2
. But

H

= T
2
, so they are not submanifolds. (In particular,
we see that compact Lie groups may have noncompact subgroups.)
Suppose H is a Lie subgroup of G, and h be the Lie algebra of H. Since
d
H
: h g is injective, we can think of h as a Lie subalgebra (= a linear subspace
that is closed under Lie bracket) of g.
Proposition 10.3. Suppose H is a Lie subgroup of G. Then as a Lie subalgebra of
g, the Lie algebra of H is
h = X g [ exp
G
(tX) H for all t R.
Proof. First suppose X h, then by naturality of exp, for any t R,
exp
G
(tX) =
H
(exp
H
(tX))
H
(H) = H.
Now suppose X , h. Consider the map
: R h G, (t, Y ) exp(tX) exp(Y ).
51
Since d exp
0
is the identity map, d
0,0
(t, Y ) = tX + Y . Since X , h, d
0,0
is
injective. It follows that there exists a small > 0 and a neighborhood U of 0 in
h such that maps (, ) U injectively into G. Shrinking U if necessary, we
may assume that exp
H
maps U dieomorphically onto a neighborhood | of e in
H. Choose a smaller neighborhood |
0
of e in H such that |
1
0
|
0
|. We pick
a countable collection h
j
[ j N H such that h
j
|
0
cover H. (This is always
possible since H is the union of countable many compact sets.)
For each j denote T
j
= t R [ exp(tX) h
j
|
0
. We claim that T
j
is a count-
able set. In fact, if [ts[ < and t, s T
j
, then exp(ts)X = exp(sX) exp(tX)
|. So exp(t s)X = exp(Y ) for a unique Y U. It follows (t s, 0) = (0, Y ).
Since is injective, we conclude Y = 0 and t = s.
Now each T
j
is a countable set. So one can nd t R such that t , T
j
for all j. It
follows that exp(tX) ,
j
h
j
|
0
= H. So X , X g [ exp
G
(tX) H for all t R.
This completes the proof.
Conversely, any Lie subalgebra gives rise to some Lie subgroup:
Theorem 10.4. If h is a Lie subalgebra of g, then there is a unique connected Lie
subgroup H of G with Lie algebra h.
Proof. Let v
1
, , v
k
be a basis of h g. Since v

i
s are left invariant vector elds
on G, linearly independent at e, they are linearly independent at all g G. In
other words, 1
g
= spanv
1
(g), , v
k
(g) gives us a k-dimensional distribution on
G. Since [v
i
, v
j
] h for all i, j, 1 is integrable. By Frobenius theorem, there is a
unique maximal connected integral manifold of 1 through e. Denote this by H.
To show that H is a subgroup, note that 1 is a left invariant distribution. So
the left translation of any integral manifold is an integral manifold. Now suppose
h
1
, h
2
H. Since h
1
= L
h
1
e HL
h
1
H, and since H is maximal, we have L
h
1
H
H. So in particular h
1
h
2
= L
h
1
h
2
H. Similarly, h
1
1
H since L
h
1
1
(h
1
) = e H
implies L
h
1
1
H H. It follows that H is a subgroup of G. Since the group operations
on H are the restriction of group operations on G, they are smooth. So H is a Lie
group.
For uniqueness, let K be another connected Lie subgroup of G with Lie algebra
h. Then K is also an integral manifold of 1. So we have K H. Since T
e
K = T
e
H
the inclusion has to be a local isomorphism. In other words, K coincide with H
near e. Since any connected Lie group is generated by any open set containing e,
we conclude that K = H.
10.2 Closed Lie Subgroups
We are more interested in those Lie subgroups that are submanifolds as well.
52
Lemma 10.5. Suppose G is a Lie group, H is a subgroup of G which is a subman-
ifold as well. Then H is closed in the sense of topology.
Proof. Since H is a submanifold of G, it is locally closed everywhere. In particular,
one can nd an open neighborhood U of e in G such that U H = U

H. Now
take any h

H. Since hU is an open neighborhood of h in G, hU H ,= . Let
h

hU H, then h
1
h

U. On the other hand, since h

H, there is a sequence
h
n
in H converging to h. It follows that the sequence h
1
n
h

H converges to h
1
h

.
In other words, h
1
h

U

H = U H. So h H, i.e.

H H. Therefore, H is
closed.
Let G be a Lie group. A subgroup H of G is said to be a closed subgroup if
it is closed in the sense of topology. Note that a closed subgroup of a Lie group is
automatically a topological group.
Lemma 10.6. Let H be a closed subgroup of a Lie group G and
h = X g [ exp(tX) H for all t R.
Then h is a Lie subalgebra of g.
Proof. Clearly h is closed under scalar multiplication. It is closed under addition
since
exp(t(X +Y )) = lim
n
(exp(
tX
n
) exp(
tY
n
))
n
and since H is closed. It is closed under Lie bracket since
exp(t[X, Y ]) = lim
n
(exp(

tX
n
) exp(

tY
n
) exp(

tX
n
) exp(

tY
n
))
n
2
.
Notice that both expressions above come from corollary 1.4 in lecture 9.
The main theorem in this section is
Theorem 10.7 (E.Cartan). Any closed subgroup H of a Lie group G is a Lie
subgroup.
Proof. Without loss of generality, we may assume H is connected. Consider the Lie
subalgebra h dened in the previous lemma. Then there is a unique connected Lie
subgroup H

of G with Lie algebra h. Let s be a complementary subspace for h in

g such that g = h s. Let U and V be two suciently small neighborhood of 0 in
h and s such that the restriction of exp
G
to U V is a dieomorphism to its image.
We Claim that Hexp
G
(U V ) = exp U. In fact, suppose exp(X+Y ) H, X U
and Y V . Then
exp(Y ) = lim
n
(exp
X +Y
n
exp
X
n
)
n
53
is in H, since H is closed. Hence Y h s = 0. Consequently, exp U is open in
both H and H

. So H =
n
(exp U)
n
= H

since both are connected.

Corollary 10.8. If : G H is Lie group homomorphism, then ker is a Lie
subgroup of G.
Corollary 10.9. Every continuous homomorphism of Lie groups is smooth.
Proof. Let : G H be a continuous homomorphism, then

= (g, (g)) [ g
G is a subgroup of GH. The projection
p :

i
GH
pr
1
G
is bijective and smooth. Thus dp is bijective near (e
G
, e
H
), which implies that p is
local dieomorphism near (e
G
, e
H
). Since it is group, p is global dieomorphism by
translation. Thus = pr
2
p
1
is smooth.
As a consequence, for any topological group G, there is at most one smooth
structure on G to make it a Lie group. (However, it is possible that one group
admits two dierent topologies and thus have dierent Lie group structures.)
54
11 The The Lie Group-Lie Algebra Correspon-
dences
11.1 Simply Connected Lie Groups
Recall that a path in M is a continuous map f : [0, 1] M. It is closed if
f(0) = f(1).
Denition 11.1. Let M be a connected Hausdor topological space.
1. Two paths f, g : [0, 1] M with the same end points (i.e. f(0) = g(0), f(1) =
g(1)) are homotopic if there is a continuous map h : [0, 1] [0, 1] M such
that h(s, 0) = f(s), h(s, 1) = g(s) for all s, and h(0, t) = f(0), h(1, t) = f(1)
for all t.
2. M is simply connected if any two paths with the same ends are homotopic.
3. A continuous surjection : X M is called a covering if each p M has a
neighborhood V whose inverse image under is a disjoint union of open sets
in X each homeomorphic with V under .
4. A simply connected covering space is called the universal cover.
For example, R
n
is simply connected, T
n
is not simply connected. The map
R
n
T
n
= R
n
/Z
n
, x x +Z
n
is a covering map. The following results are well known:
Facts from topology:
Let : X M is a covering, Z a simply connected space. Suppose : Z M
be a continuous map, such that (z
0
) = m
0
. Then for any x
0

1
(m
0
), there
is a unique lifting : Z X such that = and (z
0
) = x
0
.
Any connected manifold has a simply connected covering space.
If M is simply connected, any covering map : X M is a homeomorphism.
Theorem 11.2. The universal covering space of a connected Lie group admits a
Lie group structure such that the covering map is a Lie group homomorphism.
Proof. Let G is connected, it has a universal covering :

G G with

G simply
connected. To show

G is a Lie group, and is a Lie group homomorphism, consider
the map
:

G

G G, ( g
1
, g
2
) ( g
1
)( g
2
)
1
.
55
Choose any e
1
(e). Since

G

G is simply connected, there is a lifting map
:

G

G

G such that = and such that ( e, e) = e. Now for any
g
1
, g
2


G we dene
g
1
:= ( e, g), g
1
g
2
= ( g
1
, g
1
2
).
By uniqueness of lifting, we have g e = e g = g for all g

G, since the maps
g g e, g e g, g g
are all lifting of the map g ( g). Similarly g g
1
= g
1
g = e, and ( g
1
g
2
) g
3
=
g
1
( g
2
g
3
). So

G is a Lie group. Finally by denition ( g
1
) = ( g)
1
and ( g
1
g
2
) =
( g
1
)( g
2
). So is a Lie group homomorphism.
11.2 Lie Group Homomorphism v.s. Lie Algebra Homo-
morphism
Lemma 11.3. Suppose G, H are connected Lie groups, and : G H is a Lie
group homomorphism. If d : g h is bijective, then is a covering map.
Proof. By group invariance, it suces to check the covering property at e H.
Since d : T
e
G T
e
H is bijective, maps a neighborhood | of e in G bijectively
to a neighborhood 1 of e in H.
Let =
1
(e) G. Then is a subgroup of G. Moreover, for any a ,
L
a
(g) = (ag) = (a)(g) = (g).
So
1
(1) =
a
L
a
|. The lemma is proved if we can show L
a
1
| L
a
2
| = for
a
1
,= a
2
. We show this by contradiction. Let a = a
1
1
a
2
. If L
a
1
| L
a
2
| , = ,
then L
a
| | ,= . Consider p
2
= ap
1
L
a
| |, where p
1
, p
2
|. Then
(p
2
) = (ap
1
) = (p
1
). However, is one-to-one on |. So p
1
= p
2
. It follows
a = e, and a
1
= a
2
. Contradiction.
Corollary 11.4. Let : G H be a Lie group homomorphism with d : g h
bijective. Suppose G, H are connected and H is simply connected. Then is a Lie
group isomorphism.
Proof. Since is a covering map and H is simply connected, is homeomorphism.
In particular, and
1
are both continuous Lie group homomorphisms. By corol-
lary 2.5 in Lecture 10, is a dieomorphism.
The main theorem in this section is the following lifting property:
56
Theorem 11.5. Let G, H be Lie groups with G connected and simply connected,
g, h the Lie algebras of G, H. If : g h is a Lie algebra homomorphism, then
there is a unique Lie group homomorphism : G H such that d = .
Proof. Let k = graph() = (g, h) g h : h = (g). Obviously k is a vector
space. Let h
i
= (g
i
), i = 1, 2. Since is a Lie algebra homomorphism, we have
[h
1
, h
2
] = [(g
1
), (g
2
)] = ([g
1
, g
2
]). It follows
[(g
1
, h
1
), (g
2
, h
2
)] = ([g
1
, g
2
], [h
1
, h
2
]) = ([g
1
, g
2
], ([g
1
, g
2
])).
In other words, k is a Lie subalgebra of g h.
By the theorem we proved in lecture 10, there exists a unique connected Lie
subgroup K of G H with k as its Lie algebra. Consider the composition map
: K

K
GH
pr
1
G. This is a Lie group homomorphism, so d = dpr
1
d
K
is a
Lie algebra homomorphism. Since dpr
1
: gh g is the projection map, d : k g
is bijection. It follows that : K G is a Lie group isomorphism.
Now let : G H be the composition
G

1
K
pr
2
H.
Then this is a Lie group homomorphism with d = . This completes the proof.
Corollary 11.6. If simply connected Lie groups G and H have isomorphic Lie
algebra, then G and H are isomorphic.
11.3 Lies Fundamental Theorems
We have seen that associated to each Lie group G there is a god-given Lie algebra
g. A natural question is: To what extend will the Lie algebra determine this Lie
group? On one hand, we have seen that the Lie algebras of S
1
and R
1
are the same,
so Lie groups are not determined by its Lie algebra. On the other hand, according to
the B-C-H formula, the Lie group product near the identity e is totally determined
by the Lie bracket. So at least the Lie algebra g provides the local information for
G. The subtle relations between Lie groups and Lie algebras are described by the
following theorems observed by S. Lie at the beginning of the whole subject.
Recall that for any Lie group homomorphism f : G
1
G
2
, there is an induced
Lie algebra homomorphism df : g
1
g
2
. The assignment f df satises the
following functorial properties:
For f = Id : G G, df = Id : g g.
If f
i
: G
i
G
i+1
, i = 1, 2, are Lie group homomorphisms, then d(f
2
f
1
) =
df
2
df
1
.
57
To state the theorems, lets rst give a denition.
Denition 11.7. A local homomorphism between two Lie groups G, H is a smooth
map f from a neighborhood U of e
G
in G to a neighborhood V of e
H
in H such
that if g
1
, g
2
U and g
1
g
2
U, then f(g
1
g
2
) = f(g
1
)f(g
2
). f is called a local
isomorphism if it is a dieomorphism from U to V such that both f and f
1
are
local homomorphism.
Since the Lie algebra, as the tangent space at e, is determined by the Lie group
structure on any neighborhood of e in G. So any local homomorphism determines a
Lie algebra homomorphism as well. More over, the functorial properties above also
holds in this setting.
Theorem 11.8 (Lies rst fundamental theorem). If G
1
and G
2
are locally isomor-
phic Lie groups, then g
1
and g
2
are isomorphic Lie algebras.
Proof. Let f be the local isomorphism between G
1
and G
2
. Since df
e
: g
1
g
2
is a
Lie algebra homomorphism, We only need to show df
e
is a bijection. However, this
follows from the fact that exp is locally dieomorphism.
Conversely,
Theorem 11.9 (Lies second fundamental theorem). If g and h are isomorphic Lie
algebras, then G and H are locally isomorphic Lie groups.
Proof. Let : g h be the Lie algebra isomorphism map. As in the proof of
theorem 2.3, there exists a connected Lie subgroup K of G H whose Lie algebra
is k = (x, (x)) [ x g. We still have the facts that the composition map
: K GH
pr
1
G
is a Lie group homomorphism whose dierential d : k g is bijective. It follows
that is a local isomorphism and a dieomorphism, i.e. there is a neighborhood U
of e K, a neighborhood V of e G and a dieomorphism : V U such that
= 1
U
and = 1
V
.
Similarly the map
: K GH
pr
2
H
is also a local dieomorphism, and a local isomorphism. Now the composition
is the local isomorphism we want.
Theorem 11.10 (Lies third fundamental theorem). For any nite dimensional Lie
algebra g, there is a unique simply connected Lie group G whose Lie algebra is g.
58
Proof. According to Ados theorem (which we will not prove), every nite dimen-
sional Lie algebra is a Lie subalgebra of gl(n, R) for n large enough. So if g is a
Lie algebra, there is a connected linear Lie group G
1
whose Lie algebra is g. Let G
be the simply connected covering of G
1
. Then G is a simply connected Lie group.
Its Lie algebra is g since any covering map is a local isomorphism. The uniqueness
follows from corollary 2.4.
In fact, one can say more on Lie groups with prescribed Lie algebra. Let G be
the simply connected Lie group with Lie algebra g, and G

be a connected Lie group

with Lie algebra g. Then G

is isomorphic to a quotient of G by a discrete subgroup

of G which lies in the center of G.
59
Part III
Part III: Lie Group Actions
60
12 Lie Group Actions
12.1 Smooth Actions
Let M be a smooth manifold, Di(M) the group of dieomorphisms from M to
M.
Denition 12.1. (1) An action of a Lie group G on M is a homomorphism of
groups : G Di(M). In other words, for any g G,
g
is a map from M to M
such that
(g
1
g
2
) = (g
1
)(g
2
).
We will denote (g)(m) by g m for brevity.
(2) An action of G on M is smooth if the evaluation map
ev : GM M, (g, m) g m
is smooth.
Remarks. (1) Similarly one can dene continuous actions of topological groups on
topological spaces, or holomorphic actions of complex Lie groups on complex man-
ifolds.
(2)What we dened above is the left action. One can also dene a right action
to be an anti -homomorphism : G Di(M), i.e. such that
(g
1
g
2
) = (g
2
) (g
1
).
Any left action can be converted to a right action by requiring

g
(m) = m g := (g
1
)m = g
1
m.
Example: S
1
acts on R
2
by rotation.
Example: If X is a complete vector eld on M, then
: R Di(M), t
t
= exp(tX)
is a smooth action of R on M. In fact, every smooth action of R on M is dened
by this way.
Example: Any Lie group G acts on itself by many ways, e.g. by left multiplication,
by right multiplication and by conjugation. More generally, any Lie subgroup H of
G can act on G by left multiplication, right multiplication and conjugation.
Example: Let g be the Lie algebra of G, and g

its dual. Then the adjoint action

of G on g is
Ad : G GL(g) Di(g), g Ad
g
.
61
Similarly the coadjoint action of G on g

is
Ad

: G GL(g

) Di(g

), g Ad

g
,
where Ad

g
is dened by Ad

g
, X) = , Ad
g
1X, ) for all g

, X g.
For example, the adjoint action of GL(n, R) on M(n, R) = gl(n, R) is
Ad
C
X = CXC
1
.
A smooth manifold M together with a G-action is called a G-manifold.
Denition 12.2. Let M, N be G-manifolds. A smooth map f : M N is called
G-equivariant, or a G-map, if it commutes with the group actions, i.e.
f(g m) = g f(m)
for all g G and m M.
Example: Let G, H be Lie groups, and f : G H a Lie group homomorphism. G
acts on G itself by left translation. Dene a left G-action on H by
g h := f(g)h.
Then f is equivariant with respect to these actions.
Note that if one or both of the actions are right actions, the equivariance con-
dition above should be suitably modied.
12.2 Innitesimal Actions
Let G be a Lie group acts smoothly on M. The group Di(M) is in fact an
innite-dimensional Lie group, and : G Di(M) is a Lie group homomorphism.
So its very natural to study its linearization, i.e. its dierential as a map between
the corresponding Lie algebras. Note that a path in Di(M) is a ow on M. So as
the derivative of a path at Id, the Lie algebra of Di(M) is Vect(M). Moreover, for
any X g, the corresponding ow on M induced by is

X
: R M M, (t, m) exp(tX) m.
So the dierential of is the map
d : g Vect(M), X X
M
with
X
M
(m) =
d
dt

t=0
exp(tX) m
The map d is called the innitesimal action of g on M.
Example: Consider the left action of G on G itself. Then for any X T
e
G, X
G
is
the right invariant vector eld X
R
generated by X.
62
Lemma 12.3. Let : G Di(M) be a smooth action.
(1) For any X g, (exp tX) = exp(tX
M
).
(2) For any g G and X g, (Ad
g
X)
M
= (g
1
)

X
M
.
Proof. (1) We need to check that

m
(t) := exp(tX) m
is the integral curve of X
M
. This follows from the denition of X
M
, since
d
dt
(
m
(t)) =
d
dt
(exp tX m) =
d
ds

s=0
(exp sX exp tX m) = X
M
(exp tX m).
(2) For m M,
(Ad
g
X)
M
(m) =
d
dt

t=0
exp(tAd
g
X) m
=
d
dt

t=0
(g exp(tX)g
1
) m
= d(g)
g
1
m
d
dt

t=0
exp(tX) (g
1
m)
= d(g)
g
1
m
X
M
(g
1
m)
= (g
1
)

X
M
(m).
Proposition 12.4. The innitesimal action d is an anti-homomorphism from g
to Vect(M), i.e.
[X, Y ]
M
= [X
M
, Y
M
], X, Y g.
Proof. From Lemma 2.1 we see
(Ad
exp(tY )
X)
M
= (exp(tY ))

X
M
= exp(tY
M
)

X
M
.
In view of Theorem 3.2 in lecture 7, the derivative of the left hand side at t = 0 is
[Y, X]
M
. On the other hand, by theorem 3.4 in Lecture 3, the derivative of the right
hand side at t = 0 is [X
M
, Y
M
]. This completes the proof.
Given an action of a Lie group G on M, near e one can integrate the innitesimal
action to recover the group action. Since any connected Lie group is generated by
group elements near e, we conclude
Proposition 12.5. An action of a connected Lie group on a manifold M is uniquely
determined by its innitesimal action.
63
A natural question is, which Lie algebra anti-homomorphism can be integrate
to Lie group actions? Suppose the Lie algebra anti-homomorphism is induced by a
G-action on M, then GM decompose into submanifolds
L
m
= (g, g m) [ g G,
and each L
m
projects dieomorphically to G. So if we let L
m
be the leave con-
taining (e, m), then the point projects to g must be (g, g m). In other words, the
leaves determine the Lie group action.
Theorem 12.6 (Palais). Let G be a simply connected Lie group, : g Vect(M)
a Lie algebra anti-homomorphism such that each X
M
is complete. Then there exists
a unique action : G Di(M) such that d = .
Proof. Consider the Lie algebra anti-homomorphism
g Vect(GM), X (X
R
, X
M
).
The image of this map is an integrable distribution of dimension dimG. Let L
m
be the maximal integrable submanifold containing the point (e, m). Projecting to
the rst factor induces a smooth map
m
: L
m
G with tangent map taking
(X
R
, X
M
) to X
R
. Since the tangent map is an isomorphism, the map
m
is a local
dieomorphism. We claim that
m
is a dieomorphism.
Since G is simply connected, it is enough to show that
m
is a covering map.
Let U
0
g be a star-like neighborhood of 0 over which the exponential map is
a dieomorphism, and let U = exp(U
0
). Given (g, m

) L
m
and X U
0
, the
curve t exp(tX)g is an integral curve of X
R
. Let
X
t
be the ow of X
M
, then
it follows that t (exp(tX)g,
X
t
(m

)) is an integral curve of (X
R
, X
M
), which has
to lie in the leaf L
m
. Thus there exists an open neighborhood of (g, m

) mapping
dieomorphically onto the right translation Ug. This proves that
m
is a covering
map, and thus a dieomorphism.
Now we can dene the G action on M by
g m = pr
2
(
1
m
(g)),
where pr
2
denote the projection to the second factor map. The previous arguments
also shows that this is a Lie group action. In fact, if g =

g
i
, where g
i
= exp(X
i
),
then by property of ow we have g m = g
1
(g
2
( g
N
m) )). This completes
the proof.
64
13 Proper Actions of Lie Groups
13.1 Orbits and Stabilizers
Let : G Di(M) be a smooth action.
Denition 13.1. (1) The orbit of G through m M is
G m = g m [ g G.
(2) The stabilizer (also called the isotropic subgroup) of m M is the subgroup
G
m
= g G [ g m = m.
Remarks. (1) If m, m

lies in the same orbit, i.e. m

= g m, then
G m = G m

, so M can be decomposed into a disjoint union of orbits. We

will denote the set of orbits by M/G.
G
m
= gG
m
g
1
, i.e. G
m
and G
m
are conjugate to each other. So we can
associate to each orbit a conjugate class of subgroup in G. The conjugate
class is called the orbit type of the orbit.
(2) We will always equip with the set of orbits, M/G, the quotient topology. This
topology might be very bad in general, e.g. non-Hausdor. For example, we can
consider the natural action of R
>0
on R, then there are three orbits, +, 0, . The
open sets of the set of orbits are +, , +, 0, and the empty set, which is not
Hausdor. However, we will show that under suitable assumptions, the quotient is
a manifold.
Proposition 13.2. Let : G Di(M) be a smooth action, m M. Then
(1) The orbit G m is an immersed submanifold.
(2) The stabilizer G
m
is a Lie subgroup of G, with Lie algebra
g
m
= X g [ X
M
(m) = 0.
Proof. (1) Consider the evaluation map ev
m
: G M, g g m. It is equivariant
with respect to the left G action on G and the G action on M: ev
m
L
g
=
g
ev
m
.
Taking derivative, we get
(dev
m
)
gh
(dL
g
)
h
= (d
g
)
hm
(dev
m
)
h
.
Since (dL
g
)
h
and (d
g
)
hm
are bijective, we claim that the rank of (dev
m
)
gh
equals
the rank of (dev
m
)
h
for any g and h. It follows that the map ev
m
is of constant rank.
65
By constant rank theorem, its image, ev
m
(G) = G m, is an immersed submanifold
of M.
(2) Consider the map
ev
m
: G M, g g m,
then G
m
= ev
1
m
(m), so it is a closed set in G. It is a subgroup since is a group
homomorphism. It follows that G
m
is a Lie subgroup of G.
We have seen in Lecture 10 that the Lie algebra of G
m
is
g
m
= X g [ exp(tX) G
m
, t R.
It follows that exp(tX) m = m for X g
m
. Taking derivative at t = 0, we get
g
m
X g [ X
M
(m) = 0.
Conversely, if X
M
(m) = 0, then (t) m, t R, is an integral curve of the vector
eld X
M
passing m. It follows that exp(tX) m = (t) = m, i.e. exp(tX) G
m
for
all t R. So X g
m
.
Denition 13.3. Let Lie group G acts smoothly on M.
(1) The G-action is transitive if there is only one orbit, i.e. M = G m for any
m M.
(2) The G-action is eective (or faithful ) if
mM
G
m
= e, i.e. : G Di(M)
is into.
(3) The G-action is free if G
m
= e for all m M. It is called locally free or
innitesimally free if each G
m
is discrete.
The following properties are easy to prove:
Proposition 13.4. Let : G Di(M) be a smooth action.
(1) If the G-action is transitive, then the map g T
m
M, X X
M
(m) is onto for
all m.
(2) If the G-action is eective, then d : g Vect(M) is injective.
(3) The G-action is locally free if and only if all g
m
are 0.
13.2 Proper Actions of Lie Groups
Denition 13.5. An action of Lie group G on M is proper if the action map
GM M M, (g, m) (g m, m)
is proper, i.e. the pre-image of any compact set is compact.
66
Proposition 13.6. A smooth action of G on M is proper if and only if for every
compact subset K M, the set
G
K
= g G [ (g K) K ,=
is compact in G.
Proof. Denote the action map by F, then
G
K
= pr
1
(F
1
(K K))
where pr
1
is the projection G M G. So if the action is smooth, then for any
compact set K M, G
K
is compact.
Conversely, suppose G
K
is compact for any compact set K. If L M M is
compact, then K =
1
(L)
2
(L) M is compact. It follows that
F
1
(L) F
1
(K K) G
K
K
is a closed set in a compact set G
K
K, thus compact.
Since G
K
is always closed, we have
Corollary 13.7. If G is compact, any smooth G-action is proper.
Obviously if the G acts on M properly, then the evaluation map
ev
m
: G M, g g m
is proper. In particular, for each m M, G
m
is compact.
Theorem 13.8. Suppose G acts on M properly. Theneach orbit Gm is an embedded
closed submanifold of M, with
T
m
(G m) = X
M
(m) [ X g.
Proof. Since the evaluation map ev
m
: G M, g g m is proper, its image
ev
m
(G) = G m is closed in M.
We have already seen that G m is an immersed submanifold of M, so each
g G has a neighborhood U in G such that ev
m
(U) is an embedded submanifold of
M. Since ev
m
(U) = ev
m
(UG
m
), we may take U such that UG
m
= U. In particular,
U G
m
. To show G m is an embedded submanifold, it suces to show that there
exists a small neighborhood W of m in M such that W ev
m
(G) = W ev
m
(U).
Suppose this is not the case, then there exists a sequence of points g
k
m converging
to m with g
k
, U. The set (m, g
k
m) (m, m) is compact. So by properness,
67
the sequence (g
k
, m) is contained in a compact set. By passing to a subsequence
we may assume g
k
g

. Since g
k
m m, we see g

m = m, i.e. g

G
m
U.
Since U is open, we conclude that g
k
U for large k, contradiction.
Finally since G m is a submanifold, its the tangent space should be the image
of the tangent map. Apply this to g = e we see that, its tangent space T
m
(G m) =
(dev
m
)
e
(g) is given by the image of the tangent map g T
m
M, i.e. X
M
(m) [ X
g.
Proposition 13.9. If the G acts on M properly, the quotient M/G is Hausdor.
Proof. Suppose to the contrary, the orbits of p and q cannot be separated in M/G.
Let U
n
and V
n
be balls in M of radius 1/n around p and q respectively. Then G U
n
intersects G V
n
for all n. In other words, there exists p
n
U
n
, q
n
V
n
and g
n
G
such that p
n
= g
n
q
n
. Now the sequence (p
n
, q
n
) = (g
n
q
n
, q
n
) lies in the image of
the map
GM M M, (g, m) (g m, m)
which is proper. It follows that its limit point (p, q) also lies in the image of the
above map, i.e. p = g q for some g G. In particular, the orbits of p and q
coincide.
13.3 Proper Free Actions
Theorem 13.10. Suppose : G Di(M) is a proper free action, then the orbit
space M/G is a manifold and the quotient map : M M/G is a submersion.
Proof. Since the action is free, the map ev
m
is injective, and T
m
(G m) g. Choose
a submanifold S M with m S such that T
m
(G m) T
m
S = T
m
M. Then the
actin restricts to a smooth map : GS M whose tangent map at (e, m) equal
to the splitting g T
m
S = T
m
M. By continuity, the tangent map is still invertible
at (e, m

) for m

S close to m. By choosing S small we may assume this is the

case for all points in S. It follows from the commutative diagram (equivariance!)
GS

M

_
Lg
1
Id

_
g
1
GS

M
that : G S M has invertible tangent map everywhere, and thus is a local
dieomorphism onto its image.
We claim that this is a dieomorphism if we choose S small enough. If not,
we can choose a non-converging sequence (g
k
, m
k
) G S with m
k
m and
68
g
k
m
k
= m. So the sequence g
1
k
m = m
k
m. The set (m, m
k
) (m, m) is
compact. By properness, the sequence g
k
is contained in a compact set, thus has
a converging subsequence. However, the only limit point of the set g
k
has to be e
since g
1
k
m m and the action is free. It follows that g
k
e, contradicting with
the fact that (g
k
, m
k
) is non-converging.
Now denote by V the image of GS under in M. Then the dieomorphism
: G S V M identies V/G with S, hence gives a manifold structure
on M/G near the orbit G m. Obviously the quotient map : M M/G is a
submersion.
Notice that if H is a closed Lie subgroup in G, then the H-action on G dened
by

h
: G G, g hg
is proper and free. So we have
Corollary 13.11. Let G be a Lie group and H a closed subgroup. Then the quotient
G/H is a manifold, with the quotient map : G G/H a smooth submersion.
Proposition 13.12. Suppose G acts on M properly. Then the map
F : G/G
m
M, gG
m
g m
is a dieomorphism between the quotient G/G
m
and the orbit G m.
Proof. F is injective since if g
1
G
m
,= g
2
G
m
, then g
1
g
1
2
, G
m
, so g
1
m ,= g
2
m. So
it is a bijective map from G/G
m
to its image, G m.
Consider the evaluation map
ev
m
: G M, g g m.
Its dierential at e is given by (dev
m
)
e
(X) = X
M
(m). So
Ker(dev
m
)
e
= X g [ X
M
(m) = 0 = g
m
,
the Lie algebra of G
m
. Since the tangent space of G/G
m
at eG
m
is exactly g/g
m
, we
claim that F is immersion at eG
m
. By equivariance, it is an immersion everywhere.
If follows that F, as a map from G/G
m
to Gm, is locally dieomorphism everywhere.
Since it is bijective, it is globally dieomorphism.
69
14 The Slice Theorem
14.1 Associated Bundles
Let G be a Lie group. Recall that a right G-action on a manifold P is a Lie
group anti-homomorphism : G Di(M), i.e.
gh
=
h

g
. For example, let U
be any manifold, then

h
: U G U G, (u, g) (u, gh)
denes a right G-action on U G.
Denition 14.1. A principal G-bundle over a manifold M is a manifold P, a free
right G-action on P, and a map : P M whose level sets are exactly the orbits
of G, such that every point in M has a neighborhood U and a dieomorphism

1
(U) U G that sends
1
(m) to the ber m G and that is equivariant
with respect to the right G-action on
1
(U) and the right G-action on U G
described above.
Example: Let E M be any vector bundle of rank k. Its frame bundle is the
principal GL(k)-bundle whose ber over m M is the set of linear isomorphisms
g : R
k
E
m
, i.e. the set of basis of E
m
, with A GL(k) acting by g g A.
Proposition 14.2. Suppose : G Di(M) is a proper free action, then : M
M/G makes M into a principal G-bundle over M/G.
Proof. We dene a right G-action on M by
: G Di(M), (g)(m) := g
1
m.
Obviously this is free since is. Moreover, by the proof of theorem 3.1 in lecture
13, this makes M a principal G-bundle over G/M.
Let P M be a principal G-bundle, and let G acts linearly on a vector space
W. Then G acts on the product P W by

g
: P W P W, (p, w) (p g
1
, g w).
Denition 14.3. The associated bundle is
P
G
W := (P W)/G.
We will denote by [p, w] the equivalence class of (p, w) P W in P
G
W.
The projection map : P M induces a map P
G
W M which sends [p, w] to
(p).
70
Proposition 14.4. The associated bundle P
G
W is a vector bundle over M.
Proof. Let U M be an open set whose preimage in P is U G. Then its preimage
in P
G
W is (U G)
G
W = U W.
Remark. Every vector bundle can be obtained as an associated bundle. In fact, if
E M is a rank k vector bundle, then E = P
GL(k)
R
k
, where P is the frame
bundle dened above.
14.2 The Slice Theorem
Let G be a Lie group acts smoothly and properly on a manifold M. Recall that
for any m M, the stabilizer of m is
G
m
= g G [ g m = m.
Take the dierential map of
g
at m, we get the isotropy action of G
m
on T
m
M via

g
: T
m
M T
m
M, v d
g
(v).
Consider the orbit G m. We have seen that this is an embedded submanifold
in M, and the map
F : G/G
m
M, gG
m
g m
is a dieomorphism between the quotient G/G
m
and the orbit G m.
Obviously the tangent space T
m
(G m) is a subspace of T
m
M which is invariant
under the isotropic G
m
-action. Since the G-action on M is proper, G
m
is compact.
Hence there exists a G
m
invariant decomposition,
T
m
M = T
m
(G m) W,
where W is orthogonal to the orbit. (For example, we can x a G
m
-invariant inner
product on T
m
M and take W to be the orthogonal complement of T
m
(G m) in
T
m
M. Such an inner product exists since G
m
is compact: one can take an arbitrary
inner product and then average it over G
m
.)
Recall that G is a principal G
m
bundle over G/G
m
, since G
m
is a closed subgroup
in G. Since G
m
acts on W, if we take D be a small disc in W around the origin
with respect to some G
m
-invariant metric, G
m
also acts on D. So we can form
the associated disc bundle G
Gm
D over G/G
m
. (So locally for small open set
U G/G
m
the bundle looks like U D.)
Theorem 14.5 (The Slice Theorem). Let G be a Lie group acts smoothly and
properly on a manifold M, m M. Then there exists a G-equivariant dieomor-
phism from the disc bundle G
Gm
D onto a neighborhood of the orbit G m in
M, whose restriction to the zero section G
Gm
0 = G/G
m
is the dieomorphism
F : G/G
m
G m described above.
71
Before we prove the slice theorem, we will rst prove a special case that m is a
xed point of the G-action.
Theorem 14.6 (The Local Linearization Theorem). Let G be a compact Lie group
acting on a manifold M and let m M
G
be a xed point. Then there exists a
G-equivariant dieomorphism from a neighborhood of the origin in T
m
M onto a
neighborhood of m in M.
Proof. Let U be an invariant neighborhood of m, and let f : U T
m
M be any
smooth map whose dierential at m is the identity map on T
m
M. Consider the
average
F : U T
m
M, u F(u) =
_
G
(d
g
)
m
(f(g
1
u))dg,
where dg is the haar measure on G. (We will study the details of Haar measure
later.) Then for any g
1
G, since d(g
1
g) = dg,
F(g
1
u) =
_
G
(d
g
)
m
f(g
1
g
1
u)dg =
_
G
d
g
1
g
f(g
1
u)d(g
1
g) = d
g
1
F(u).
In other words, F is equivariant with respect to the isotropy G-action on T
m
M and
the given G action on U. Moreover, since
d((d
g
)
m
f
1
g
)
m
= (d
g
)
m
df
m
(d
1
g
)
m
= Id
for all g G, we claim that dF
m
is the identity map. So by inverse function theorem,
F is dieomorphism near m.
Proof of the slice theorem: Since the action is proper, the stabilizer G
m
is
compact. By the local linearization theorem above, there exists a G
m
-equivariant
dieomorphism from a neighborhood of 0 in T
m
M to a neighborhood of m in M
such that (0) = m. Take a small disc D with respect to some G
m
invariant inner
product on W as described above. Consider the map
: G
Gm
D M, [g, v] g (v).
This is well-dened for D small enough contained in the domain of , since if
(g
1
, v
1
) (g
2
, v
2
), then there is some g G
m
such that g
2
= g
1
g
1
and v
2
= g v
1
.
So
g
2
(v
2
) = g
1
g
1
(g v
1
) = g
1
(v
1
).
At [e, 0], the dierential of is the composition of T
[e,0]
(U D) = T
m
(G m) D
T
m
M with d. Since is dieomorphism, is a local dieomorphism at [e, 0].
By G-equivariance, it is a local dieomorphism at all points of the form [g, 0]. It
remains to show that is bijective for D small enough.
72
Assume to the contrary that there exists u
n
, v
n
0 in W and g
n
, h
n
G
such that [g
n
, u
n
] ,= [h
n
, v
n
] while g
n
(u
n
) = h
n
(v
n
). Without loss of gen-
erality, we may assume h
n
= e. Then g
n
(u
n
) = (v
n
) m. Since the ac-
tion is proper, and under the map G M M M the sequence (g
n
, (u
n
))
goes to the convergent sequence ((v
n
), (u
n
)), there is a converging subsequence
g
n
i
g

. Contradicts with the fact that is a local dieomorphism at [a

, 0].

14.3 Applications
In this section we always assume G acts on M smoothly and properly.
Application to xed point sets.
Proposition 14.7. For any non-trivial subgroup H G, its xed point set
M
H
= m M [ g m = m for all g H
is a disjoint union of closed submanifolds of M.
Proof. Obviously M
H
is closed in M for any H. Observe that the xed point set of
H coincides with the xed point set of its closure

H, and moreover, for any m M
H
,

H G
m
. So without loss of generality, we may assume that H is a compact Lie
subgroup of G.
Let F be a connected component of M
H
, and m F be a point. By the
local linearization theorem, there exists a neighborhood U of m in M and an H-
equivariant dieomorphism of U with an open subset of the vector space W =
T
m
M. This dieomorphism carries U F to W
H
, a linear subspace consisting
of those vectors that are xed by H. It follows that F is a submanifold.
Corollary 14.8. For any vector X in the Lie algebra g, the zero set
M
X
= m M [ X
M
(m) = 0
is a disjoint union of closed submanifolds of M.
Proof. Let H = exp(tX) [ t R. Then M
X
= M
H
.
Application to Eectiveness.
Proposition 14.9 (Eective Locally Eective). Suppose M is connected and the
G-action is eective and proper. Let U M be any invariant open subset. Then
the G-action on U is also eective.
73
Proof. We need to show that the group
K = g G [ g u = u for all u U
is trivial. This is a closed subgroup of G, and its xed point set M
K
contains the
open set U. Take the connected component N of M
K
that contains U, then by Prop.
3.1, N is a closed submanifold of M. However, dimN = dimM since N contains
an open set. It follows that N is an open submanifold of M. In other words, N is
both open and closed in the connected manifold M. It follows that N = M, and
thus M
K
= M. Since the action is eective, we conclude K = e.
Other Applications.
Includes
Orbifolds as quotient of locally free proper actions.
Existence of G-invariant partition of unity for G-invariant covering.
Existence of G-invariant geometric structures, includes G-invariant Rieman-
nian metric, G-invariant connection etc.
74
15 Orbit Types
15.1 Homogeneous Spaces
Let Lie group G acts on M smoothly. In general there will be many G-orbits in
M. However, if the group action is transitive, there will be only one G-orbit the
manifold M itself is an orbit.
Denition 15.1. Let G be a Lie group. A homogeneous G-space, or a homogeneous
space, is a manifold M endowed with a transitive smooth G-action.
Example: The natural action of O(n) on S
n1
R
n
is transitive. (Gram-Schmidt
procedure.) So S
n1
is a homogeneous space.
Example: Let H G be a closed subgroup. Then the left G-action on the quotient
manifold G/H is transitive. It follows that G/H is a homogeneous G-space. Note
that the stabilizer of the left G-action at any point in G/H is always conjugate to
H.
It turns out that the last example above is the universal example.
Theorem 15.2. Let M be a homogeneous space of the Lie group G, m M. Then
the map
F
m
: G/G
m
M, gG
m
g m
is an equivariant dieomorphism.
Proof. It is easy to check that F
m
is well-dened, bijective, and is equivariant
with respect to the left G-action on G/G
m
and the G-action on M. Since G
acts transitively on G/G
m
, the equivariance implies that F
m
is immersion. Since
dimG/G
m
= dimF
m
, the dierential map of F
m
is not only injective, but actually
bijective. It follows that F
m
is locally dieomorphism. Being locally dieomorphism
and globally bijective, F
m
is a dieomorphism.
It follows that to study homogeneous spaces of a Lie group G, it suces to study
its closed subgroups. One can take G/H as denitions of homogeneous spaces.
Example: Consider again the natural O(n) action on S
n1
. If we choose the base
point in S
n1
to be the north pole N = (0, , 0, 1), it is easy to see that the
isotropy group G
N
= O(n 1). We conclude
S
n1
O(n)/O(n 1).
Example: Let k < n. The manifold of all k-dimensional linear subspaces in R
n
is the Grassmann manifold Gr
k
(R
n
). Obviously O(n) acts transitively on Gr
k
(R
n
),
and the isotropy group of

R
k
is O(k) O(n k). It follows that
Gr
k
(R
n
) O(n)/(O(k) O(n k)).
75
Note that in the special case k = 1, Gr
1
(R
n
) = RP
n1
is the real projective space.
Similar the complex Grassmann manifolds and the complex projective spaces are
homogeneous U(n)-spaces.
15.2 Orbit Types
Suppose G acts on M smoothly. Recall that for any m M and g G, the
stabilizer subgroups G
m
and G
gm
are related by
G
gm
= gG
m
g
1
.
In other words, to each orbit G m there corresponds a conjugate class of subgroups
in G.
Denition 15.3. The orbit type of a point m M (with respect to the given
G-action) is the conjugacy class of its stabilizer G
m
in G.
Using the stabilizer group G
m
, one can rewrite the H-xed point set as
M
H
= m M [ H G
m
.
Similarly one dene for any closed subgroup H G,
M
H
= m M [ H = G
m

M
(H)
= m M [ H is conjugate to G
m

M
(H)
= m M [ H is conjugate to a subgroup of G
m
.
Obviously M
(H)
is the ow-out of M
H
under the G-action: M
(H)
= G M
H
. So in
particular M
(H)
is G-invariant. Similarly M
(H)
, as the ow out of M
H
under the
G-action, is G-invariant. We are mainly interested in M
(H)
, since by denition it is
the set of points with orbit type (H).
Example: Consider the rotation action of S
1
on M = S
2
. Then the set of xed
points M
S
1
consists of two points: the north pole and the south pole. There are
exactly two orbit types for this action: the xed points have orbit type H = S
1
, and
other points have the trivial orbit type H = e.
Proposition 15.4. Suppose G acts on M properly. Then for any closed subgroup
H G, the set M
(H)
is a submanifold of M.
Proof. It is enough to prove that the orbit of each point with stabilizer H has a
neighborhood whose intersection with M
(H)
is a submanifold. According to the
76
slice theorem, we may assume M = G
H
W. Since M
(H)
is G-invariant, a point
[g, w] M
(H)
if and only if [e, w] M
(H)
. Notice that
g G
[e,w]
(g, w) (e, w)
h H s.t. gh = e, and h
1
w = w
g H and w M
G
[e,w]
,
So as a subgroup of G, G
[e,w]
is conjugate to H if and only if G
[e,w]
equals H and
that w W
H
. It follows M
(H)
= G
H
W
H
, which is a vector sub-bundle, and thus
a submanifold.
So we can decompose M into submanifolds
M =
_
(H)
M
(H)
,
or decompose it into the connected components of these sets. In general these
submanifolds are not connected, and their components do not all have the same
dimension. The following proposition shows that the decomposition above is locally
nite:
Proposition 15.5. If the G-action on M is proper, the number of orbit types is
locally nite. More explicitly, every point in M has a neighborhood that intersects
only a nite number of the sets M
(H)
.
Proof. We apply induction to n = dimM. If n = 0, the proposition is obvious,
since M is discrete in this case. Now we suppose the proposition is proved for all
the G-manifolds of dimension less or equal to n 1.
Assume dimM = n. According to the slice theorem, it suces to show that
in the model G
H
W, with a compact Lie group H, the number of orbit types is
nite. Obviously the orbit type of the point [g, 0] is (H) for any g. Now consider
the points [g, w] for w ,= 0. Take a H-invariant metric on W. Since the G-action on
W is linear, we have
(G
[g,w]
) = (G
[e,w]
) = (G
[e,w/w]
),
which equals the G-conjugacy class of the stabilizer of w/|w| under the H-action
on the unit sphere in W. By induction hypothesis, the H-action on the unit sphere
in W has only a nite number of orbit types. Since any two conjugate subgroups in
H are automatically conjugate as subgroups of G, the proposition follows.
Corollary 15.6. If M is compact, only a nite number of orbit types can occur.
Theorem 15.7. Suppose Lie group G acts on M properly, and M is connected.
Then there exists a closed subgroup H = H
prin
of G such that the set M
(H)
is dense
and open in M.
77
Denition 15.8. The class (H) = (H
prin
) is called the principal orbit type of the
given G-action.
Proof. We still prove the result by induction on n = dimM. If n = 0 and M is
connected, there is only one orbit and the theorem is obvious. Now assume the
theorem is proved for G-manifolds of dimension less than or equal to n 1.
Suppose M is a G-manifold of dimension n. We rst prove the theorem for
the model G
H
W. By induction hypothesis, the H-action on the unit sphere
S = S
n1
W has a principal orbit type K, i.e. S
(K)
is open and dense in S. Since
(G
H
W)
(K)
G
H
(R

S
(K)
),
we conclude that (G
H
W)
(K)
, as a submanifold that contains an open dense subset,
must itself be an open dense subset.
Now consider the general case. The previous argument shows that a neighbor-
hood of every orbit contains a principal orbit type. Suppose U and U

are two open

sets with non-empty intersection, with M
(K)
U dense and open in U, and M
(K

)
U

dense and open in U

, then M
(K)
and M
(K

)
must intersect, and thus equal. If follows
from connectivity that all these principal orbit types must be the same.
78
Part IV
Part IV: Representations of Lie
Groups
79
16 Elementary Representation Theory
16.1 Representations of Lie Groups and Lie Algebras
We have studied the Lie group actions on smooth manifolds. A very important
case is that the manifolds are linear vector spaces, and moreover the Lie group
actions are linear:
Denition 16.1. A representation of a Lie group G is a pair (V, ), where V is a
vector space, and : G GL(V ) is a linear G-action on V .
Remarks. (1) Although a representation contains a pair (V, ), we will always use
V or to denote the representation. Also we will write g v instead of (g)v for
simplicity.
(2) In this denition we did not specify whether G and V are real or complex.
Usual we will take V to be a complex vector space, even if G is real Lie group.
(In general complex representations are much simpler than real ones, since complex
matrices theory are much easier than real matrices theory.)
(3) In this denition V might be innite dimensional, e.g. a Hilbert space. But
we will only concentrate on nite dimensional representations in this course. The
dimension of V is called the dimension of this representation.
Examples: The following are some examples of representations.
The trivial representation: G is arbitrary, V = C, and (g) = Id for any
g G.
The standard representation for linear Lie groups: G is a linear Lie group, and
is the inclusion map : G GL(n).
The adjoint representation: G is arbitrary, V = g, and (g) = Ad(g).
The coadjoint representation: G is arbitrary, V = g

, and (g) = Ad(g)

.
Let G acts on a smooth manifold M, and F(M) be the space of complex-valued
functions on M. Then the action induces a representation of G on F(M) by
g (x) = (g
1
x).
In particular if M = G and the G-action is the left action, we get the left
regular representation of G.
Denition 16.2. Let V, W be two representations of G. A morphism (or an inter-
twining map ) between them is a linear equivariant map f : V W, i.e. f is linear
and
f(g v) = g f(v)
80
for all g G and v V . The set of all such morphisms is denoted by Hom
G
(V, W).
As usual, an invertible morphism is called an isomorphism, and isomorphic repre-
sentations are always called equivalent representations.
One of the central problems in algebra is to classify all the objects with the same
algebraic structure. In representation theory we have
Basic problem: Classify all representations of a given Lie group G, up to isomor-
phism.
The Lie algebra representations are dened by the same way:
Denition 16.3. A representation of a Lie algebra g is a pair (V, ), where V
is a vector space and : g gl(V ) a Lie algebra morphism. (In algebra a g-
representation V is also called a g-module.)
Similarly one can dene morphisms of representations of Lie algebra. The set
of g-morphisms between V and W is denoted by Hom
g
(V, W).
According to Lie group-Lie algebra correspondence, we have
Proposition 16.4. Let G be a Lie group with Lie algebra g.
(1) Every representation : G GL(V ) of G denes a representation = d : g
gl(V ) of g. Moreover, any morphism between representations of G is automatically
a morphism between the corresponding representations of g.
(2) If G is connected and simply connected, then any representation of g can be
uniquely lifted to a representation of G. Moreover, Hom
G
(V, W) = Hom
g
(V, W).
Example: The dierential of the adjoint representation G on g gives the adjoint
representation of g on g: for any X g, ad(X) End(g) maps Y to [X, Y ].
16.2 Operations on Representations
It is well known that given several vector spaces, one can produce many other
vector spaces using algebraic operations such as direct sum. Similarly given several
representations, one can produce new representations using these algebraic opera-
tions.
Subrepresentations and Quotient representations
Denition 16.5. Let V be a representation of G. A subrepresentation of V is a
G-invariant linear subspace W V together with the restriction of to W.
Suppose W is a subrepresentation of V . Then since W is a linear subspace of
V , one can form the quotient space V/W. It follows from the invariance of W that
the G-action on V descends to an action on V/W by
g (v +W) = g v +W.
81
This gives a representation of G on the quotient V/W, and is called the quotient
representation of V under W.
Similarly one can dene the subrepresentations and quotient representations of
a given g-representation. It is easy to check that if G is a connected Lie group, and
V a representation of G, then W V is a subrepresentation of V if and only if it is
a subrepresentation of g.
Direct Sum and Tensor Product
Recall that if V, W are vector spaces of dimension n and m respectively, with
basis v
1
, , v
n
and w
1
, , w
m
, then
V W is a vector space of dimension n + m. A basis of V W is given by
v
1
, , v
n
, w
1
, , w
n
.
V W is a vector space of dimension nm. A basis of V W is given by
v
i
w
j
: 1 i n, 1 j m. In general,
(

a
i
v
i
) (

b
j
w
j
) =

a
i
b
j
v
i
w
j
.
The dual V

is a vector space of dimension n. A basis of V

is given by the
dual basis v

i
, where v

i
(v
j
) =
i
j
.
The set of linear maps Hom(V, W) is a vector space of dimension mn. In fact,
Hom(V, W) = W V

.
Now we will extend these operations to representations. Suppose V and W be
two representations of G or g, then
The direct sum V W is a representation of G or g: The G-action on V W
is given by
g (v, w) := (g v, g w).
Similarly one can dene the g-action on V W.
The tensor product V W is a representation of G if we dene
g (v w) := g v g w.
However, the g-action on V W is tricker. For example, one cannot dene
the g-action to be X v w = X v X w as before, since this is not even
linear in X. Note that the g-action is the dierential of the G-action, we have
X(v w) =
d
dt
[
t=0
(e
tX
v e
tX
w) = (X v) w +v (X w).
82
The dual representation V

of G is dened such that

g v

= (g
1
)
T
v

,
where for any linear map A : V V , A
T
is the adjoint map A
T
: V

, i.e.
if we denote the natural pairing between V and V

by , ), then Av, w

) =
v, A
T
w

). Note that for any v V and w

, we have
g v, g w

) = v, w

).
Similarly the g-action on V

is dened by
X v

= X
T
v

,
and we have for any v V, w

,
X v, w

) +v, X w

) = 0.
It follows that the space of linear maps Hom(V, W) = W V

admits a
natural structure as a representation of G and g. More explicitly, the G-action
on Hom(V, W) is given by
(g f)(v) = g (f(g
1
v)),
and its dierential is
(X f)(v) = X f(v) f(X v).
16.3 Irreducible Representations
To classify all the representations of a given Lie group G, it is natural to study
the simplest possible representations.
Denition 16.6. Let (V, ) be a representation of Lie group G. V is irreducible (or
simple) if it has no subrepresentations other than 0 and V .
For example, the standard representation of O(n) on R
n
is irreducible. Obviously
any one dimensional representation is irreducible.
If a representation is not irreducible, then it has a nontrivial subrepresentation
W, which give us a short exact sequence
0 W V V/W 0
of representations. A natural question is: whether this exact sequence splits, i.e.
whether V = W V/W. If this is the case, then W and V/W are built-up blocks
of the representation V , which are simpler then V since the dimensions are lower.
83
Denition 16.7. A representation V is called completely reducible (or semisimple)
if it is isomorphic to a direct sum of irreducible representations.
Modied Basic Problems:
(1) Classify all the irreducible representations of a given Lie group G.
(2) Decompose any completely reducible representation into direct sum of irre-
ducibles.
(3) Determine for which G all representations are completely irreducible.
Denition 16.8. A representation (V, ) of G is unitary if V admits a G-invariant
positive-denite Hermitian inner product, i.e. (g) is unitary for any g G.
Proposition 16.9. Any unitary representation is completely reducible.
Proof. Let V be any reducible representation of G, and W V a subrepresentation.
Pick an invariant inner product on V , then W

is invariant under the G-action. It

follows that W

is also a subrepresentation, and V = W W

. We can do this
procedure again and again until each component is irreducible.
Theorem 16.10. Any representation of a compact Lie group admits an invariant
inner product.
Proof. Let V be a representation of G. Start with any inner product on V , we can
dene a new inner product via
v, w)
new
:=
_
G
g v, g w)dg,
where dg is the Haar measure on G. It is easy to check that , )
new
is invariant
under the G-action:
g v, g w)
new
= v, w)
new
.
It follows that any representation of a compact Lie group is unitary.
Corollary 16.11. Any representation of a compact Lie group is completely re-
ducible.
84
17 The Haar Measure
17.1 Density
First lets recall the change of variables formula from calculus: if f is an inte-
grable function dened in domain U R
n
, : U

U a bijective smooth map from

U

to U, with x = (x

), then
_
U
f(x)dx =
_
U

f((x

))

(x
1
, , x
n
)
(x

1
, , x

n
)

dx

,
where
(x
1
, ,xn)
(x

1
, ,x

n
)
is the Jacobian matrix of the coordinate change x

x = (x

).
Denition 17.1. Let V be a vector space of dimension n. A density on V is a map
: V
n
C satisfying
(Av
1
, , Av
n
) = [ det A[ (v
1
, , v
n
)
for any v
i
V and A End(V ).
Remarks. (1) An n-form on V is a map : V
n
C such that
(Av
1
, , Av
n
) = (det A) (v
1
, , v
n
).
So if
n
(V ) is a n-form, [[ is a density.
(2) Similarly one can dene, for any C, an -density by requiring
(Av
1
, , Av
n
) = [ det A[

(v
1
, , v
n
).
Proposition 17.2. The set of densities on V , [[(V ), is a one dimensional vector
space.
Proof. Linearity of [[(V ) follows from denition. Since [[ [[(V ) for any

n
(V ), it is at least one dimensional.
To show it is exactly one dimensional, lets suppose
1
(e
1
, , e
n
) =
2
(e
1
, , e
n
)
for some basis e
1
, , e
n
of V . Then for any v
1
, , v
n
, one can choose a unique
linear map A on V that sends e
i
to v
i
. It follows that

1
(v
1
, , v
n
) = [ det A[
1
(e
1
, , e
n
) = [ det A[
2
(e
1
, , e
n
) =
2
(v
1
, , v
n
).
Thus
1
=
2
if they coincide on one basis. Now we take any nonzero element

n
(V ). Denote a = [[(e
1
, , e
n
). If [[(V ) and (e
1
, , e
n
) = b, we
must have = a/b [[. This completes the proof.
85
A density is positive if it takes value in [0, +) and is not identically zero. We
will denote the set of positive densities on V by [[
+
(V ). For example, if
n
(V ),
then [[ is a positive density on V , and in general any density is a multiple of [[.
Now suppose M is a smooth manifold of dimension n. A density on M assign
to each point m M a density on T
m
M. Suppose (U, x
1
, , x
n
) is a coordinate
patch near m M, then we can write any density on U as
(x) = f(x)[dx
1
dx
n
[
for some function f on U. A density is continuous or smooth if for any coordinate
patch, f(x) is continuous or smooth.
We can pull back densities as follows: If f : M N is a smooth map, and is
a density on N, then f

, the pull-back of , is a density on M dened by

(f

)
m
(X
1
, , X
n
) =
f(m)
(df
m
(X
1
), , df
m
(X
n
)).
Check: f

is a density on M.
Now we dene the integration of compactly supported continuous densities on
M.
Step 1. First suppose is a compactly supported continuous density on R
n
. Then
we can write = f[dx
1
dx
n
[ for some continuous function f support on a
compact set D R
n
. Dene
_
R
n
:=
_
R
n
f(x)dx
1
dx
n
=
_
D
f(x)dx
1
dx
n
.
To dene the integration of densities on manifolds, we need the following
Lemma 17.3. Suppose U, V are open sets in R
n
, and : U V is a dieomor-
phism, is a density on V , then
_
V
=
_
U

.
Proof. Denote = f[dx
1
dx
n
[, then

= f((x)) [det d[ [dx

1
dx
n
[,
and the lemma follows from the change of variable formula.
Step 2. Secondly suppose is a continuous density on M supported on a coordinate
chart (, U, V ), we dene
_
U
:=
_
V
(
1
)

.
86
This is well-dened, since if ( ,

U,

V ) is another coordinate chart and is also
supported in

U, then
_

V
(
1
)

=
_
V
(
1
)

(
1
)

=
_
V
(
1
)

,
where we used the fact that
1
is a dieomorphism from (U

U) to (U

U),
and that (
1
)

= (
1
)

.
Step 3. Finally suppose is any compactly supported continuous density on M.
Take a nite open cover U
i
of support of by coordinate charts, then U
i
, U
0
=
M U
i
is a nite cover of M. The partition of unity theorem claims that there
exists smooth functions
i
supported in U
i
satisfying 0
i
1 and

i
1.
Now we can dene _
M
=

_
U
i

i
.
It is not hard to check that this is independent of choices of open cover, and choices
of partition of unity, so the integration of compactly supported densities are well
dened. Moreover, it satises the following propositions:
Proposition 17.4. Let , be compactly supported densities on M.
(1) (Linearity)
_
M
(a +b) = a
_
M
+b
_
M
.
(2) (Positivity) If is a positive density,
_
M
> 0.
(3) (Dieomorphism Invariance) If : N M is a dieomorphism, then
_
M
=
_
N

.
17.2 The Haar Measure
Now let G be a Lie group.
Denition 17.5. A density on G is called left invariant if L

g
= for all g G.
Similarly one can dene the right invariant density.
Theorem 17.6. Left invariant density exists on any Lie group G, and is unique up
to a multiplicative constant.
Proof. Take any linearly independent left invariant vector elds X
1
, , X
n
on G.
Denote the dual co-vector elds by X

1
, , X

n
. Then = X

1
X

n
is a nonzero
left invariant n-form on G. It follows that = [[ is a left invariant density on G.
The uniqueness follows from the fact that [[(T
e
G) is one dimensional, and each
left invariant density is determined by its value (e) at e. Thus if
1
,
2
are two
left invariant densities on G, and
1
(e) = C
2
(e), then
1
= C
2
everywhere.
87
Any density on G denes a measure on G via
f C
c
(G)
_
G
f.
We will call a left invariant density a left Haar measure. It is normalized, denoted
by = dg, if it satises the normalizing condition
_
G
dg = 1. Note that the left
invariance means
d(hg) = dg,
or equivalently,
_
G
f(hg)dg =
_
G
f(g)dg
for any xed h G.
Theorem 17.7. If G is a compact Lie group, there exists a unique normalized left
Haar measure.
Proof. In the proof of the previous theorem, we actually proved the existence of
positive left invariant density on G. Let be a positive left invariant density on
G, then
_
G
= C is a positive real number (not innity since G is compact). Now
dg = /C is the left Haar measure on G.
17.3 Modular Function
Similarly one can dene the right Haar measure, and show the existence and
uniqueness of normalized right Haar measure on G. In general a left Haar measure
will not be a right Haar measure.
Now let G be a Lie group and a positive left Haar measure on G. Since left
multiplication commutes with right multiplication, R

g
is another positive left Haar
measure on G. Thus there exists a positive constant, (g), such that = (g)R

g
.
Denition 17.8. The function : G R
+
is called the modular function of G. G
is called unimodular if (g) 1 for any g G.
Note that by denition, a Lie group is unimodular if and only if every left
Haar measure is also a right Haar measure. So we can speak of Haar measure on
unimodular Lie groups, without indicating left or right. Obviously any commutative
Lie group is unimodular.
Proposition 17.9. The modular function : G R
+
is a Lie group homomor-
phism.
88
Proof. By denition
= (g
1
g
2
)R

g
1
g
2
= (g
1
g
2
)R

g
2
R

g
1
.
On the other hand, we have
R

g
1
= (g
2
)R

g
2
R

g
1
,
and thus
= (g
1
)R

g
1
= (g
1
)(g
2
)R

g
2
R

g
1
.
It follows (g
1
g
2
) = (g
1
)(g
2
).
Theorem 17.10. Any compact Lie group is unimodular.
Proof. If G is compact, the image (G) of G is a compact subgroup of R
+
. However,
the only compact subgroup of R
+
is 1. The theorem follows.
Proposition 17.11. If G is unimodular, the inversion map g g
1
is an isometry.
Proof. Let be a left Haar measure on G. Denote the inversion map by . Then it is
easy to check that

is a right Haar measure. Since G is unimodular,

= C for
some positive constant C. Since is the identity map, we must have C = 1.
89
18 Schurs Orthogonality
18.1 Schurs Lemma
Lemma 18.1 (Schurs Lemma). Let V , W be irreducible representations of Lie
group G.
(1) If f : V W is an intertwining map, then either f 0, or f is invertible.
(2) If f
1
, f
2
: V W are two intertwining maps and f
2
,= 0 , then there exists
C such that f
1
= f
2
.
Proof. (1) Suppose f is not identically zero. Since ker(f) is an invariant subset in
V , it must be 0. So f is injective. In particular, f(V ) is a nonzero subspace of
W. On the other hand, it is easy to check that f(V ) is an invariant subspace of W.
It follows that f(V ) = W, and thus f is invertible.
(2) Since f
2
,= 0, it is invertible. So f = f
1
2
f
1
is an intertwining map from
V to V itself. Let be one of the eigenvalues of the linear map f. Then f Id
is an intertwining operator from V to V which is not invertible. It follows that
f Id 0, and thus f
1
= f
2
.
Note that in the proof we showed in particular
Corollary 18.2. (1) Let V be an irreducible representation of G, and f : V V
is an intertwining map of V with itself. Then f = Id for some C.
(2) Conversely, if (, V ) is a unitary representation of G, and Hom
G
(V, V ) =
CId, then (, V ) is an irreducible representation.
Proof. (1) This is a direct consequence of the previous lemma.
(2) Let 0 ,= W V be a G-invariant subspace. We need to show that W = V .
Let P : V W be the orthogonal projection (with respect to the given G-invariant
inner product). Since both W and W

are G-invariant, we have for any g G,

(g)P = (g) = P(g) on W, and (g)P = 0 = P(g) on W

. It follows that
P : V W V is an intertwining operator, and so P = Id for some C. Now
P
2
= P implies = 1, and thus W = V .
Recall that the center Z(G) of a Lie group G is
Z(G) = h G : gh = hg, g G.
Corollary 18.3. If (, V ) is an irreducible representation of G, then for any h
Z(G), (h) = Id for some C.
90
Proof. Suppose h Z(G), then for any g G,
(h)(g) = (hg) = (gh) = (g)(h).
In other words, (h) : V V is an intertwining operator, and the conclusion
follows.
18.2 Schur Orthogonality for Matrix Coecients
Let (V, ) be a representation of a Lie group G. If we choose a basis e
1
, , e
n
of V , we can identify V with C
n
, and represent any g G by matrices:
(g)v =
_
_
_

11
(g)
1n
(g)
.
.
.
.
.
.
.
.
.

n1
(g)
nn
(g)
_
_
_
_
_
_
v
1
.
.
.
v
n
_
_
_
for v =

v
i
e
i
. So if we take L
j
: V C be the function
L
j
(

i
v
i
e
i
) = v
j
,
then
ij
is the function on G given by

ij
(g) = L
i
((g)e
j
).
Denition 18.4. For any v V, L V

, the map
: G C, (g) = L((g)v)
is called a matrix coecient of G.
Obviously any matrix coecients of G is a continuous function on G. In fact,
they form a subring of C(G):
Proposition 18.5. If
1
,
2
are matrix coecients for G, so are
1
+
2
and
1

2
.
Proof. Let (
i
, V
i
) be representations of G, v
i
V
i
, L
i
V

i
such that
i
(g) =
L
i
(
i
(g)v
i
). Then (
1

2
, V
1
V
2
) is a representation of G, L
1
L
2
V

1
V

2
=
(V
1
V
2
)

and
(L
1
L
2
)((
1

2
)(g)(v
1
, v
2
)) =
1
(g) +
2
(g).
Similarly we have a linear functional L
1
L
2
on V
1
V
2
satisfying (L
1
L
2
)(v
1
v
2
) =
L
1
(v
1
)L
2
(v
2
), and thus
(L
1
L
2
)((
1

2
)(g)(v
1
v
2
)) =
1
(g)
2
(g).
91
Now suppose G is a compact Lie group, and dg the normalized Haar measure
on G. Recall that L
2
(G), the space of square-integrable functions with respect to
this Haar measure, is the completion of the space of continuous functions on G with
respect to the inner product
f
1
, f
2
)
L
2 =
_
G
f
1
(g)f
2
(g)dg.
Theorem 18.6 (Schurs Orthogonality I). Let (
1
, V
1
) and (
2
, V
2
) are two non-
isomorphic irreducible representations of a compact Lie group G. Then every matrix
coecient of
1
is orthogonal in L
2
(G) to every matrix coecient of
2
.
Proof. Fix G-invariant inner products on V
1
and V
2
respectively. Suppose

i
(g) =
i
(g)v
i
, w
i
), i = 1, 2
are matrix coecients for
i
, where v
i
, w
i
V
i
. Fix a basis of V
1
such that e
1
= v
1
.
Dene a linear map f : V
1
V
2
by f(e
1
) = v
2
and f(e
k
) = 0 for all k 2. Consider
the map
F : V
1
V
2
, v F(v) =
_
G

2
(g)f(
1
(g
1
)v)dg.
F is linear since f is. It is also equivariant, since
F(
1
(h
1
)v) =
_
G

2
(g)f(
1
(hg)
1
v)dg =
2
(h
1
)
_
G

2
(hg)f(
1
(hg)
1
v)dg =
2
(h
1
)F(v).
By Schurs lemma, F(v) = 0 for any v, and in particular, F(v), w
2
) = 0. On the
other hand, for any j,

2
(g)f(
1
(g
1
)e
j
) =
2
(g)f(

1
(g
1
)
kj
e
k
) =
1
(g
1
)
1j

2
(g)(v
2
),
where
1
(g
1
)
kj
=
1
(g
1
)e
j
, e
k
) is the matrix coecients of
1
with respect to the
basis e
1
, , e
n
. It follows that
_
G

1
(g
1
)e
j
, e
1
)
2
(g)v
2
, w
2
)dg = 0
for any j. So by linearity,
_
G

1
(g
1
)w
1
, v
1
)
2
(g)v
2
, w
2
)dg = 0.
Note that the inner product is G-invariant,

1
(g
1
)w
1
, v
1
) = w
1
,
1
(g)v
1
) =
1
(g)v
1
, w
1
) =
1
(g).
So the theorem follows.
92
Theorem 18.7 (Schurs Orthogonality II). Let (, V ) be an irreducible representa-
tion of a compact Lie group G, with invariant inner product , ). Then
_
G
(g)w
1
, v
1
)(g)w
2
, v
2
)dg =
1
dimV
w
1
, w
2
)v
1
, v
2
).
Proof. Dene the linear maps f, F : V V as above. Then F is equivariant, and
thus F = Id for some = (v
1
, v
2
) C. On the other hand, when we take

1
=
2
= , the computation in the previous proof shows
(v
1
, v
2
)w
1
, w
2
) = F(w
1
), w
2
) =
_
G
(g
1
)w
1
, v
1
)(g)v
2
, w
2
)dg.
Since the Haar measure in invariant under the inversion map g g
1
, the right
hand side is invariant if we exchange w
1
with v
2
, and exchange w
2
with v
1
. It follows
that (v
1
, v
2
) = Cv
2
, v
1
) for some constant C. Finally if we take v
1
= v
2
= e
1
a
unit vector, then since Tr((g)f(g
1
)) = Tr(f) = 1, and thus Tr(F) = 1, we must
have C =
1
dimV
.
18.3 Schur Orthogonality for Characters
Let (V, ) be a representation of Lie group G.
Denition 18.8. The character of V is the function on the group dened by

V
: G C,
V
(g) = Tr[
V
((g)).
It is easy to check that the character of a representation satises

V
1
V
2
=
V
1
+
V
2
.

V
1
V
2
=
V
1

V
2
.

V
=
V
.

V
(ghg
1
) =
V
(h).
Note that the character
V
(g) =

ii
(g) is a linear combination of matrix
coecients. So as a corollary of the Schurs orthogonality for matrix coecients, we
get
Theorem 18.9 (Schurs Orthogonality for Characters).
(1) If (, V ) is an irreducible representation of a compact Lie group G, then
_
G
[
V
(g)[
2
dg = 1.
93
(2) If (
1
, V
1
) and (
2
, V
2
) are non-isomorphic irreducible representations of a
compact Lie group G, then
_
G

V
1
(g)
V
2
(g)dg = 0.
As a corollary, we can show that the character determines the representation:
Corollary 18.10. Two representations (, V ) and (, W) of a compact Lie group
G are isomorphic if and only if
V
=
W
.
Proof. Since G is compact, any representation is completely reducible. We decom-
pose V and W into irreducible representations
V =

i
m
i
V
i
, W =

i
n
i
V
i
,
where V
i
s are non-isomorphic irreducible representations of V , and m
i
, n
i
are non-
negative integers. Then obviously
V
=

m
i

V
i
and
W
=

n
i

V
i
. Suppose

V
=
W
, then we must have
m
i
=
V
,
V
i
) =
W
,
V
i
) = n
i
.
It follows that V is isomorphic to W.
Remark. In particular we see that for any completely reducible representation, there
is a unique way to decompose it into irreducible ones.
Corollary 18.11. A representation (, V ) is irreducible if and only if
_
G
[
V
(g)[
2
dg =
1.
Proof. Let V =

n
i
V
i
be the decomposition as above, then
_
G
[
V
(g)[
2
dg =

n
2
i
.
94
19 The Peter-Weyl Theorem
19.1 Some Functional Analysis
Let H be a (complex) Hilbert space, i.e. a (nite or innite dimensional) vector
space with an inner product, such that H is complete with respect to the induced
metric [v[ = v, v)
1/2
. A linear operator T : H H is said to be bounded if there
exists C > 0 such that
[Tv[ C[v[, v H.
Denition 19.1. Let H be a Hilbert space, and T : H H a bounded operator.
(1) T is self-adjoint if for any v, w H, Tv, w) = v, Tw).
(2) T is compact if for any bounded sequence v
1
, v
2
, in H, the sequence Tv
1
, Tv
2
,
has a convergent subsequence.
We say that C is an eigenvalue of an operator T on H, with eigenvector
0 ,= v H, if Tv = v. We will denote the set of all eigenvalues of T by Spec(T),
and the eigenspace for Spec(T) by H

. Self-adjoint operator on Hilbert space is

the (innite dimensional) generalization of Hermitian operators acting on C
n
. For
example, we have
Lemma 19.2. If T is self-adjoint, then Spec(T) R, and for ,= Spec(T),
H

.
Proof. If Spec(T) with eigenvector v, and T is self-adjoint, then
v, v) = Tv, v) = v, Tv) = v, v) =

v, v).
So =

, i.e. R.
Similarly if ,= Spec(T), 0 ,= v H

and 0 ,= w H

, then
v, w) = Tv, w) = v, Tw) = v, w) = v, w).
So v, w) = 0, i.e. v w.
The following spectral theorem for compact self-adjoint operators in Hilbert
space will play a crucial rule in the proof of Peter-Weyl theorem.
Theorem 19.3 (The Spectral Theorem). Let T be a compact self-adjoint operator
on a Hilbert space H, and denote by N(T) = Ker(T) the null space of T. Then
(a) The set of nonzero eigenvalues is a countable set (and thus discrete).
95
(b) If the set Spec(T) =
1
,
2
, is not a nite set, then lim
k

k
= 0.
(c) For each eigenvalue 0 ,= Spec(T), dimH

< .
(d) Denote by v
(k)
1
, , v
(k)
n(k)
an orthonormal basis of H
k
, where n(k) = dimH
k
.
Then
v
(k)
j
[ 1 j n(k), k = 1, 2,
form an orthonormal basis of N(T)

.
19.2 Convolution on Compact Lie Groups
Let G be a compact Lie group with the normalized Haar measure dg, and C(G)
the ring of continuous functions on G. Since G has volume 1, we have inequalities
|f|
1
|f|
2
|f|

for any f C(G), where | |

p
is the is the L
p
norm, i.e.
|f|
p
=
__
G
[f(g)[
p
dg
_
1/p
,
and
|f|

= sup
gG
[f(g)[.
It follows that C(G) L

(G) L
2
(G) L
1
(G). Recall that L
2
(G) is a Hilbert
space, with inner product
f
1
, f
2
) =
_
G
f
1
(g)f
2
(g)dg.
Denition 19.4. For any f
1
, f
2
C(G), we can dene the convolution by
(f
1
f
2
)(g) :=
_
G
f
1
(gh
1
)f
2
(h)dh.
Remarks. (1) Use the change of variable h h
1
g one can check
(f
1
f
2
)(g) =
_
G
f
1
(h)f
2
(h
1
g)dh.
(2) We will use convolution for f
1
C(G) and f
2
L
1
(G).
Now for any C(G), consider the operator
T

: L
1
(G) L
1
(G), f f.
96
Proposition 19.5. For any C(G) and f L
1
(G), T

(f) C(G), and

|T

(f)|

||

|f|
1
.
In particular, the operator T

is a bounded operator on L
2
(G).
Proof. Since is uniformly continuous, there exits a neighborhood U of e such that
[(g) (kg)[ < for all g G and k U. It follows that for any g, g

G with
g

g
1
U,
[T

f(g) T

f(g

)[ = [
_
G
((gh
1
) (g

h
1
))f(h)dh[

_
G
[(gh
1
) (g

g
1
gh
1
)[[f(h)[dh[ |f|
1
.
So T

f is a continuous function. Moreover,

|T

(f)|

= sup
gG
[
_
G
(gh
1
)f(h)dh[ ||

_
G
[f(h)[dh = ||

|f|
1
.
The main result in this section is
Proposition 19.6. Let C(G). Then
(1) The operator T

is a compact operator on L
2
(G).
(2) If (g
1
) = (g), then T

is self-adjoint on L
2
(G).
Before proving this lets rst remind you the Ascoli-Arzela theorem. Recall that
a subset U C(X) is equicontinuous if for any x X and > 0, there exists a
neighborhood N of x such that [f(x) f(y)[ < holds for all y N and all f U.
Ascoli-Arzela Theorem. Let X be compact, U C(X) a bounded and equicontinu-
ous subset. Then every sequence in U has a uniformly convergent subsequence.
Proof of Proposition 19.6. (1) It suces to show that any sequence in
B = T

(f) [ f L
2
(G), |f|
2
1
has a convergent subsequence. According to the fact |f|
2
|f|

and the Arzela-

Ascoli theorem, it suces to show that B is bounded and equicontinuous in (C(G), |
|

). The boundedness follows from the previous proposition. Now lets show the
equicontinuity: Since G is compact, is uniformly continuous, i.e. > 0, there
97
exists a neighborhood U of e G such that [(kg) (g)[ < for all g G and
k U. So if |f|
2
1,
[( f)(kg) ( f)(g)[ = [
_
G
((kgh
1
) (gh
1
))f(h)dh[

_
G
[(kgh
1
) (gh
1
)[ [f(h)[dh
|f|
1

holds for all k N and all f L
2
(G) with |f|
2
1.
(2) We have by denition
T

(f
1
), f
2
) =
_
G
_
G
(gh
1
)f
1
(h)f
2
(g)dgdh
while
f
1
, T

(f
2
)) =
_
G
_
G
(hg
1
)f
1
(h)f
2
(g)dgdh.
So if (g
1
) = (g), T

is self-adjoint.
19.3 The Peter-Weyl Theorem
The right translation R(g) of G on C(G), dened by
(R(g)f)(x) = f(xg),
is a linear G-action on C(G). Suppose C(G), and Spec(T

). I claim that
the eigenspace H

is G-invariant. In fact,
(T

R(g)f)(x) =
_
G
(xh
1
)f(hg)dh =
_
G
(xgh
1
)f(h)dh = R(g)(T

f)(x) = R(g)f(x).
Lemma 19.7. A function f C(G) is a matrix coecient on G if and only if the
functions R(g)f span a nite dimensional vector space.
Proof. It is easy to check that if f C(G)

for some nite dimensional representa-

tion , then R(g)f C(G)

for any g. It follows that R(g)fs span a vector space

of dimension no more than n
2
, where n = dim.
Conversely, suppose the functions R(g)f span a nite dimensional vector space
V , then (R, V ) is a nite dimensional representation of G, and if we dene a func-
tional L : V C by L() = (e), it is clear that L(R(g)f) = f(g), so f is a matrix
coecient of G.
98
Theorem 19.8 (Peter-Weyl Theorem). Let G be a compact Lie group. Then the
matrix coecients of G are dense in C(G).
Proof. Let f C(G) be a continuous function on G. Since G is compact, f is
uniformly continuous, i.e. there exists a neighborhood U of e in G such that for any
g U and h G,
[f(g
1
h) f(h)[ < /2.
Now let be a nonnegative function supported in U such that
_
G
(g)dg = 1 and
(g) = (g
1
). It follows that T

is a self-adjoint compact operator. Then for any

h G,
[(T

f)(h)f(h)[ = [
_
G
((g)f(g
1
h)(g)f(h))dg[
_
U
(g)[f(g
1
h)f(h)[dg

2
.
On the other hand, by construction of , T

is compact and self-adjoint on L

2
(G). It
follows by the spectral theorem that for all Spec(T

), H

are nite dimensional

for ,= 0, mutually orthogonal, and span L
2
(G). So if we write f =

, where
f

, then

|f

|
2
2
= |f|
2
2
< . In particular, one can nd such that

0<||<
|f

|
2
2
<

2||

.
Now let

f = T

||
f

). Then

f

||
H

, and moreover for any g G,

R(g)

f

||
H

. Since

||
H

is a nite dimensional vector space, we con-

clude that

f is a matrix coecient of G. Note that
T

(f)

f = T

(f
0
+

0<||<
f

) = T

0<||<
f

),
so
|T

(f)

f|

||

0<||<
f

|
2
<

2
.
It follows
|f

f|

|f T

f|

+|T

f

f|

< .
Since C(G) is dense in L
2
(G), we immediately get
Corollary 19.9. Let G be a compact Lie group. Then the matrix coecients of G
are dense in L
2
(G).
99
20 Applications of the Peter-Weyl Theorem
20.1 Faithful Representations
Denition 20.1. A representation (, V ) is faithful if as a linear action, is eec-
tive, i.e. the group homomorphism : G GL(V ) is injective.
Lemma 20.2. Let G be a compact Lie group. Then for any g ,= e, there exists a
representation (, V ) such that (g) ,= Id.
Proof. Take any function f C(G) such that f(e) = 0 and f(g) = 1. Then
there is a representation (, V ) of G and a matrix coecient C(G)

such that
|f |

<
1
3
. So in particular (e) ,= (g). Since (x) = L((x)v) for some xed
v V and L V

, we must have (g) ,= (e) = Id.

Theorem 20.3. Any compact Lie group possesses a faithful representation.
Proof. According to the previous lemma, for any g
1
G
0
, g
1
,= e, there exists
a representation (
1
, V
1
) of G such that
1
(g
1
) ,= Id. Thus Ker(
1
) is a closed
subgroup of G, thus a Lie group by itself. Moreover, since G
0
Ker(
1
), we
must have dimKer(
1
) < dimG. If dimKer(
1
) ,= 0, we do this procedure again,
i.e. take a g
2
Ker(
1
)
0
and a representation (
2
, V
2
) such that
2
(g
2
) ,= Id. It
follows that Ker(
1

2
) is a compact Lie subgroup in G with dimKer(
1

2
) <
dimKer(
1
). Continuing this procedure, we will get a sequence of representations
(
i
, V
i
), 1 i N, of G, such that dimKer(
1

N
) = 0. Since G is compact,
Ker(
1

N
) = h
1
, , h
M
is a nite set. Now we choose representations
(
i
, W
i
), 1 i M such that
i
(h
i
) ,= Id. It follows that the representations

1

N

1

M
is a faithful representation of G.
As a corollary, we are ready to show that any compact Lie group is a linear Lie
group:
Corollary 20.4. Any compact Lie group is isomorphic to a closed subgroup of U(N)
for N large.
Proof. Take a faithful representation (, V ) of G. Since G is compact, there exists
a G-invariant inner product on V . It follows : G GL(V ) maps G into U(V )
U(N). The injectivity of implies that G is isomorphic to its image, a closed
subgroup of U(N).
Remark. There exists noncompact Lie groups that does not admit any nite dimen-
sional faithful representation, and thus are not linear Lie groups. One example of
such Lie groups is the metaplectic group Mp
2n
, the double cover of the symplectic
100
group Sp
2n
. Although it doesnt have any nite dimensional faithful representation,
it does have faithful innite dimensional representations, such as the Weil repre-
sentation. (This is another reason why compact Lie groups are much simpler than
noncompact Lie groups: for compact Lie groups there is no innite dimensional
irreducible representations.)
20.2 Class Functions
Denition 20.5. A function f C(G) is called a class function if it is conjugate
invariant, i.e. f(gxg
1
) = f(x) for all x, g G.
For example, let (, V ) be a representation of G, then the character
V
is a class
function on G. Recall that two representations are isomorphic if and only if they
have the same character. We will call the characters of irreducible representations
the irreducible characters.
Theorem 20.6 (Peter-Weyl Theorem for Class Functions). The irreducible charac-
ters generate a dense subspace of the space of continuous class functions.
Proof. Suppose is a class function. For any > 0, by the Peter-Weyl theorem,
one can nd a representation (, V ) of G and a matric coecient f C(G)

such
that | f|

< . Consider the function dened on G by (x) =

_
G
f(gxg
1
)dg.
Obviously is a class function, and | |

< . We claim that is a linear

combination of irreducible characters, thus proves the theorem.
In fact, since G is compact, (, V ) =

i
(
i
, V
i
) is a direct sum of nitely
many irreducible representations. Since f C(G)

C(G)

i
, we can write
f(g) =

i
L
i
(
i
(g)v
i
) for some v
i
V
i
and L
i
V

i
. So
(x) =

i
L
i
__
G

i
(g)
i
(x)
i
(g
1
)v
i
dg
_
.
We have already shown in the proof of Schurs orthogonality that the map
v
i

_
G

i
(g)
i
(x)
i
(g
1
)v
i
dg
as the average of
i
(x), is a linear equivariant map on V
i
. So
_
G

i
(g)
i
(x)
i
(g
1
)v
i
dg = (x)v
i
for some (x) C. Computing the trace as we did before, we conclude
(x) =
1
dimV
i
Tr(
i
(x)) =
1
dimV
i

V
i
(x).
101
It follows that
(x) =

1
dimV
i
L
i
(v
i
)
V
i
(x)
is a linear combination of irreducible characters.
Corollary 20.7. For any class function f in L
2
(G),
f =

IrrRep(G)
f,

as an L
2
function with respect to L
2
-convergence, and
|f|
2
2
=

IrrRep(G)
f,

)
2
.
20.3 Irreducible Representation of SU(2)
Recall that
SU(2) = U(2) SL(2, C) =
_
g =
_

_
[ , C, [[
2
+[[
2
= 1
_
.
It is not hard to see that SU(2) is a compact Lie group, and any g SU(2) is
conjugate to a diagonal matrix of the form
e(t) =
_
e
it
0
0 e
it
_
.
Here are some known representations of SU(2):
The trivial representation of SU(2) on C: (
0
, V
0
= C)
The standard representation of SU(2) on C
2
: (
1
, V
1
= C
2
)
The previous representation induces an representation of SU(2) on C(G), given
by
(g f)(z
1
, z
2
) = f(z g) = f(z
1
+z
2
,

z
1
+ z
2
).
The previous representation is too large. There is a much nicer subrepresen-
tation: the representation space being the space of polynomials on C
2
. (If
f P(C
2
) is a polynomial, then g f is again a polynomial.) This is still an
innite dimensional representation.
102
The previous representation has an important invariant: the degree of the
polynomial. It follows that for any n 0, the space of homogeneous polyno-
mials of degree n on C
2
,
V
n
= Spanz
k
1
z
nk
2
[ 0 k n
is a representation of SU(2). Moreover, dimV
n
= n +1. Note that the SU(2)
action on V
n
is given by
(g P)(z) = P(g
1
z) = P(z
1
+z
2
,

z
1
+ z
2
)
for g SU(2), P V
n
.
Proposition 20.8. The representations (
n
, V
n
) described above are irreducible.
Proof. According to Corollary 1.2(2) of lecture 18, it is enough to show that any
linear intertwining map A Hom
SU(2)
(V
n
, V
n
) is a scalar.
For any 0 k n, we denote by p
k
(z) the monomial z
k
1
z
nk
2
. Then

n
(e(t))p
k
= e
i(2kn)t
p
k
.
Fix a t R such that all the eigenvalues e
i(2kn)t
are dierent. Since A is equivariant,
A commutes with
n
(e(t)). It follows that A preserves all the eigenspaces of
n
(e(t)).
In other words, there exits
k
, 0 k n, such that
Ap
k
=
k
p
k
.
Let E
0
be the eigenspace of A with eigenvalue
0
. Then
0
E
0
, and E
0
is SU(2)-
invariant since A is an intertwining map. So if we denote by r() the rotation
matrix
r() =
_
cos sin
sin cos
_
SU(2),
then
n
(r())p
0
E
0
. So

n
(r())p
0
(z
1
, z
2
) = (cos z
1
+ sin z
2
)
n
=

k
_
n
k
_
(cos )
k
(sin )
nk
p
k
(z
1
, z
2
) E
0
.
It follows that p
k
E
0
for all 0 k n, i.e. E
0
= V
n
. Thus A is a scalar
multiplication.
In the proof above we see that the eigenvalues of
n
(e(t)) are exactly e
i(n2k)t
for 0 k n. It follows that the character

Vn
(e(t)) =
n

k=0
e
i(2kn)t
=
sin(n + 1)t
sin t
103
Theorem 20.9. The representations (
n
, V
n
) described above are the only irreducible
representations of SU(2).
Proof. According to Peter-Weyl theorem and the Schurs orthogonality for charac-
ters, it is enough to show that the linear span of the irreducible characters
Span
Vn
, n = 0, 1,
is dense in the space of continuous class functions. Since any matrix g SU(2) is
conjugate to some e(t), any class function on SU(2) is uniquely determined by its
restriction on
T = e(t) [ t R,
which is isomorphic to S
1
. Moreover, since e(t) is conjugate to e(t), any class
function restricted to T can be identied with an even function on S
1
. On the other
hand, since

Vn
(e(t))
V
n1
(e(t)) = e
int
+e
int
= 2 cos(nt),
we have
Span
Vn
, n = 0, 1, = Spancos(nt), n = 0, 1, 2, .
The basic theory on Fourier series tells us that Spancos(nt), n = 0, 1, 2, is
dense in the space of even functions on S
1
. It follows that Span
Vn
, n = 0, 1,
is dense in the space of continuous class functions.
104
21 Abelian Lie Groups: Structure and Represen-
tations
21.1 Structure of Abelian Lie Groups
Denition 21.1. (1) A Lie group G is abelian if g
1
g
2
= g
2
g
1
for any g
1
, g
2
G.
(2) A Lie algebra g is abelian if [X, Y ] = 0 for any X, Y g.
Lemma 21.2. A connected Lie group G is abelian if and only if its Lie algebra g is
abelian.
Proof. Suppose G is abelian, then for any X, Y g,
(t) := exp(tX) exp(tY )
is a one-parameter subgroup with

(0) = X +Y . It follows
exp X exp Y = exp(X +Y ).
So if we think of g as a additive Lie group, exp is a Lie group homeomorphism. On
the other hand, exp is also a local dieomorphism, thus as Lie group g is locally
isomorphic to G. So the Lie groups G and g have the same Lie algebra, and g =
Lie(G) = Lie(g) is abelian.
Conversely, suppose g is abelian, then by the Baker-Campbell-Hausdor for-
mula,
exp X exp Y = exp(X +Y ).
for X, Y close to 0 g. It follows that G is abelian near e. Since G is connected, it
has to be abelian globally.
Recall that exp is a local dieomorphism, and any neighborhood of e generate
the identity component G
0
. So any g G
0
has the form
g = exp(X
1
) exp(X
N
)
for X
1
, , X
N
g close to 0. In general one cannot expect that exp is a surjective
map to G
0
. However, if G is abelian, then
exp(X
1
) exp(X
N
) = exp(X
1
+ +X
N
),
so we have
Corollary 21.3. If G is a connected abelian Lie group, then exp : g G is onto.
105
Corollary 21.4. Let G be a connected Lie group of dim 1, then either G R or
G S
1
.
Proof. The only one dimensional Lie algebra is the abelian Lie algebra R. So G is
an abelian Lie group, the exponential map exp is onto and is a Lie group homo-
morphism. Let K = ker(exp), then K is a Lie subgroup of R. Since exp is locally
isomorphism, K is an isolated subgroup of R. If K = 0, then G = R. Other-
wise let a be the smallest nonzero number in K, then K = aZ, and G = R/aZ is
isomorphic to the circle.
Obviously the products
T
k
R
l
= S
1
S
1
R R
is abelian. Conversely,
Theorem 21.5. Any connected Lie group G has the form T
k
R
l
.
Proof. Since exp is a Lie group homomorphism, and is locally dieomorphism near
0, K = ker(exp) which is a discrete subgroup of g. Let U be the R-vector space
generated by K, and V be a complement of U in g. Since exp is onto,
G g/K = (U V )/K U/K V,
where U/K T
k
is a torus and V R
l
for some l.
Corollary 21.6. Any compact connected Abelian Lie group is a torus.
21.2 Representation of the Torus
Lemma 21.7. Any irreducible representation of an abelian Lie group G is one
dimensional.
Proof. Let (, V ) is an irreducible representation of G. Then (g)(h) = (h)(g)
for any g, h G. So (g) : V V is an intertwining operator. According to Schurs
lemma, (g) = I is a scalar multiplication. It follows that any subspace of V is
invariant. Since V is irreducible, we must have dimV = 1.
Now lets restrict ourselves to the representations of the compact abelian Lie
groups T
k
. Note that any one dimensional representation is irreducible, so to nd
out all irreducible representations of T
k
is equivalent to nd out all one dimensional
representations of T
k
. For any n = (n
1
, , n
k
) Z
k
, there is a one-dimensional
representation
n
of T
k
on C, given by
(e
i
1
, , e
i
k
) z = e
i(n
1

1
++n
k

k
)
z.
Note that if n
1
= = n
k
= 0, this gives the trivial representation.
106
Theorem 21.8. The representations
n
given above are pair-wise dierent, and give
all the irreducible representations of T
k
.
Proof. The character for the representation
n
is e
i(n
1

1
++n
k

k
)
. Suppose (n
1
, , n
k
) ,=
(n

1
, , n

k
), then there is an j such that n
j
,= n

j
, and
_
2
0
e
i(n
j
n

j
)
d = 0.
so
_
T
k
e
i(n
1

1
++n
k

k
)
e
i(n

1
++n

k
)
d
1
2

d
k
2
=
k

l=1
_
2
0
e
i(n
l
n

l
)
l
d
l
2
= 0.
So
n
,=
n
for n ,= n

.
On the other hand, by classical Fourier analysis, the characters
e
i(n
1

1
++n
k

k
)
[ (n
1
, , n
k
) Z
k

span L
2
(T
k
). So there is no irreducible character that is orthogonal to all the
characters of
n
s. In other words, there is no other irreducible representations.
Remark. We can formulate the representations of a general connected abelian Lie
group G without identify it with T
n
, as follows: Let G be a connected abelian Lie
group. The group lattice of G is the lattice Z
G
= K = ker(exp) g, and the weight
lattice is the dual
Z

G
= g

[ () 2Z for Z
G
.
Then the irreducible representations of G are parametrized by Z

G
. In other words,
any Z

G
gives an irreducible representation

of G on C by
exp(X) z = e
i(X)
z.
Corollary 21.9. Any representation of a connected Abelian Lie group G is of the
form
(e
X
)(z
1
, , z
n
) = (e
i
1
(X)
z
1
, , e
in(X)
z
n
)
for some weights
1
, ,
n
Z

G
.
107
21.3 Fourier Theory
We have seen that the completeness of the characters of T
k
is equivalent to the
classical Fourier theory. In the same spirit one can dene the Fourier theory on
general compact abelian Lie groups, and even general compact Lie groups. We will
denote by

G the set of irreducible representations of G.
Lets start with compact abelian Lie groups G. For any function f L
2
(G), we
can dene its Fourier transform

f :

G C by

f() = f,

)
L
2.
We equip with

G the counting measure. The inner product on
l
2
(

G) = :

G C [

G
[()[
2
<
is given by
, )
l
2 =

G
()().
Proposition 21.10. Let G be a compact abelian Lie group.
(1) (Completeness) The set of characters

[

G form a complete orthonormal
system for L
2
(G).
(2) (Inversion formula) For any f L
2
(G), f =

f()

.
(3) (The Plancherels formula) For f L
2
(G), |f|
L
2 = |

f|
l
2.
Proof. Since G is abelian, any function on G is a class function. Now (1) comes
from the Peter-Weyl theorem for class functions (theorem 2.2 in lecture 20), and
(2), (3) follows from corollary 2.3 in lecture 20.
Remark. For G = T
n
, we have seen

G = Z
n
, and for any f L
2
(G),

f(m
1
, , m
n
) =
1
(2)
n
_
2
0

_
2
0
f(
1
, ,
n
)e
i(m
1

1
++mnn)
d
1
d
n
.
Now suppose G is a connected compact Lie group in general. As above one can
dene the scalar valued Fourier transform by
f L
2
(G)

f() = Tr( (f))
for (, V )

G, where
: L
2
(G) End(V ), ( (f))(v) =
_
G
f(g)g vdg.
108
Note that

f() =
_
G
f(g)
V
(g)dg = f,
V
)
2
= f,
V
).
According to the Peter-Weyl theorem for class functions, we have
Proposition 21.11. Let G be a compact Lie group. Then the map f (

f())

G
establishes an isomorphism from L
2
class functions on G to l
2
(

G).
To establish an isomorphism from the whole space L
2
(G), one has to the operator-
valued Fourier transform instead of scalar-valued transform. In this case the substi-
tution for l
2
(

G) for nonabelian groups is
Op(

G) =

(,V )

G
End(V ).
(For abelian group G, the right hand side is exactly l
2
(

G).)
Denition 21.12. Let G be a compact Lie group. The operator valued Fourier
transform T : L
2
(G) Op(

G) is
Tf =
_
(dimV )
1
2
(f)
_
(,V )

G
.
Theorem 21.13. Let G be a compact Lie group. Then
(1) The inverse transform of T is given by
(T

)
(,V )

(,V )

G
(dimV )
1/2
Tr(T

(g
1
)).
(2) (Plancherels formula) For any f L
2
(G), |f|
2
2
= |Tf|
2
Op(

G)
=

(,V )

G
dimV | (f)|
2
2
.
109
Part V
Part V: Structure of Compact Lie
Groups
110
22 Maximal Tori
22.1 Maximal Tori
By a torus we mean a compact connected abelian Lie group, so a torus is a Lie
group that is isomorphic to T
n
= R
n
/Z
n
.
Denition 22.1. Let G be a compact connected Lie group. A subgroup T G is
a maximal torus if T is a torus and there is no other torus T

with T T

G.
Example: Consider G = U(n). Then the subgroup T of diagonal matrices in U(n)
is clearly isomorphic to T
n
. It is actually a maximal torus, for if there is a strictly
larger one, then one can nd some element g of U(n) that commutes with all elements
of T. But T contains diagonal matrices with n distinct eigenvalues, and any matrix
commutes with such matrices must be diagonal, so we get a contradiction. Note
that in this example, any g U(n) is conjugate to some element in this maximal
torus. In other words, we have
U(n) =
_
gU(n)
gTg
1
.
Similar, we can dene
Denition 22.2. Let g be the Lie algebra of a compact Lie group. Then any
maximal abelian subalgebra of g is called a Cartan subalgebra of g.
Proposition 22.3. Let G be a compact Lie group. Then there is a one-to-one
correspondence between maximal tori in G and Cartan subalgebras in g.
Proof. We have seen in the previous lecture that a connected subgroup T G is
abelian if and only if its Lie algebra t is abelian. So the proposition follows.
Note that any one dimensional subspace of g is abelian, it is clear that maximal
abelian subalgebra of g exists. It follows that maximal tori always exist. In fact,
Corollary 22.4. Let G be a compact Lie group. Then any torus in G is contained
in a maximal torus.
Proof. Let T
1
G be a torus. Then its Lie algebra t
1
= Lie(T
1
) is an abelian Lie
subalgebra of g. Since dimg is nite, one can always nd a Cartan subalgebra t of
g containing t
1
. The corresponding Lie subgroup T is obviously a maximal torus in
G that contains T
1
.
111
Remark. (1) A maximal torus is the same as a maximal connected abelian Lie
subgroups, since the closure of any connected abelian Lie subgroup is a torus.
(2) Any g G closed to e has the form exp(X) for some X g, thus sits in
the connected abelian Lie subgroup exp(tX). So any g G in a neighborhood of
e sits in some maximal torus. We will show later that exp is onto for any compact
connected Lie group. As a result, any g G sits in some maximal torus.
22.2 Cartan Subalgebras
Let G be a compact Lie group, and T G a torus in G. Consider the conjugation
action of G on itself,
c(g) : G G, h ghg
1
.
Obviously c(g)(T) = gTg
1
is again a torus in G for any g G. Moreover, conjuga-
tion action preserves the inclusion relation. It follows that if T is a maximal torus,
so is gTg
1
. Similarly if t g is a (maximal) abelian Lie subalgebra, then for any
g G, Ad(g)(t) is a (maximal) abelian Lie subalgebra.
Recall that for any X g, the centralizer of X in g by
Z
g
(X) = ker(ad(X)) = Y g [ [X, Y ] = 0.
Lemma 22.5. Let G be a compact Lie group and t a Cartan subgroup of g. Then
there exists X t such that t = Z
g
(X).
Proof. Since t is abelian, t ker(ad(X)) for any X t. In particular, if X
1
, , X
n

is a basis of t, then t
i
ker(ad(X
i
)). In fact we have t =
i
ker(ad(X
i
)), otherwise
we can take a vector

X
i
ker(ad(X
i
)) t, then

t = spanX
1
, , X
n
,

X is an
abelian subalgebra of g which is strictly larger than t, a contradiction. In what
follows we will show
ker(ad(X
1
)) ker(ad(X
2
)) = ker(ad(X
1
+tX
2
))
for some t R, and thus by induction, t = ker(X
1
+ t
1
X
2
+ + t
n1
X
n
) for some
t
i
R, completing the proof.
Take an inner product on g that is invariant under the adjoint G-action on g,
i.e.
Ad(g)Y
1
, Ad(g)Y
2
) = Y
1
, Y
2
)
for any g G and Y
1
, Y
2
g. Taking derivative, it follows that for any X g,
ad(X)Y
1
, Y
2
) = Y
1
, ad(X)Y
2
),
112
i.e. ad(X) is skew-symmetric. Let t
X
= ker(ad(X)) and u
X
= Im(ad(X)). Then it
follows that u
X
(t
X
)

, and by dimension counting, u

X
= (t
X
)

. In particular u
X
is an ad(X)-invariant subspace of g.
Now suppose X
1
, X
2
t. Obviously t
X
1
t
X
2
t
X
1
+X
2
. If u
X
1
u
X
2
= 0,
then t
X
1
t
X
2
= t
X
1
+X
2
, otherwise there is a Y g such that ad(X
1
+ X
2
)(Y ) = 0
while ad(X
1
)(Y ) = ad(X
2
)(Y ) ,= 0, a contradiction.
Finally we suppose u
X
1
u
X
2
,= 0. Since X
1
, X
2
commute, the Jacobi identity
implies that ad(X
1
) and ad(X
2
) commute. So ad(X
2
) preserves the ad(X
1
)-invariant
subspaces u
X
1
. As a consequence, the intersection u
X
1
u
X
2
is invariant under
ad(X
1
+ tX
2
) for any t R. Let p(t) = det ad(X
1
+ tX
2
)[
u
X
1
u
X
2
. Then p(t)
is a polynomial in t, and is not identically zero since p(0) ,= 0. It follows that
there exists a t
1
,= 0 such that p(t
1
) ,= 0, i.e. ad(X
1
+ t
1
X
2
) is invertible on
u
X
1
u
X
2
. This implies t
X
1
t
X
2
= t
X
1
+t
1
X
2
, for otherwise there exists Y g such
that 0 ,= [X
1
, Y ] = t
1
[X
2
, Y ] u
X
1
u
X
2
, and thus
ad(X
1
+t
1
(X
2
))([X
1
, Y ]) = [X
1
+t
1
X
2
, [X
1
, Y ]] = [X
1
, [Y, X
1
+t
1
X
2
]] = 0.
Contradiction with the fact that ad(X
1
+t
1
(X
2
)) is invertible on u
X
1
u
X
2
.
So any Cartan subalgebra is the centralizer of some element in g. Such elements
X g are called regular elements. Obviously if X is a regular element, so is tX for
t ,= 0. It turns out that the set of regular elements form an open dense subset in g.
Theorem 22.6. Let G be a compact Lie group, and t a Cartan subalgebra. Then
for any X g, there exists g G such that Ad(g)X t.
Proof. According to the previous lemma, there exits Y g such that t = ker(ad(Y )).
Consider the continuous function on G dened by
f(g) = Y, Ad(g)X),
where , ) is an Ad-invariant inner product on g. Since G is compact and f is
continuous, f attains its maximum at some g
0
G. It follows that for any Z g,
the function
t Y, Ad(exp(tZ))Ad(g
0
)X)
has a maximum at t = 0. Taking derivative at t = 0, we get
Y, ad(Z)(Ad(g
0
)X)) = 0,
i.e.
Y, ad(Ad(g
0
)X)(Z)) = 0
for all Z g. Since ad(Z) is skew-symmetric, we have
ad(Ad(g
0
)X)(Y ), Z) = 0
for all Z g. So ad(Ad(g
0
)X)(Y ) = 0, i.e. Ad(g
0
)(X) ker(ad(Y )) = t.
113
Corollary 22.7. Let G be a compact Lie group, and t g a Cartan subalgebra.
Then any Cartan subalgebra t

of g is of the form Ad(g)(t) for some g G.

Proof. Suppose t = ker(ad(X)) and t

= ker(ad(X

)). Then there exists g G such

that Ad(g)X

t. It follows
Ad(g)(t

) = Ad(g)Y [ [Y, X

] = 0
= Ad(g)Y [ [Ad(g)Y, Ad(g)X

] = 0 = Z
g
(Ad(g)X

).
Since Ad(g)X

t and t is abelian, we have Ad(g)(t

) t. But t is maximal abelian,

so Ad(g)(t

) = t.
Corollary 22.8. Let G be a compact Lie group and T G a maximal torus. Then
any maximal torus of G is of the form gTg
1
.
Proof. Suppose T

is a maximal torus in G, then t

is a Cartan subalgebra of g. By
the previous lemma, t

= Ad(g)(t) for some g G. It follows

gTg
1
= c(g)(T) = c(g) exp(t) = exp(Ad(g)(t)) = exp(t

) = T

.
In particular, we see that any two maximal tori in a compact Lie group have
the same dimension.
Denition 22.9. The rank of a compact Lie group G is the dimension of its maximal
tori.
114
23 Cartans Theorem
23.1 Surjectivity of the exponential map
We know that in general the exponential map exp : g G is not surjective,
even if G is connected. In lecture 21 we showed that exp is surjective if G is abelian
and connected. The main theorem in this section is
Theorem 23.1. Let G be a compact connected Lie group. Then the exponential
map exp is surjective.
Proof. We prove this by induction on the dimension of G. If dimG = 1, then we
have seen that G must be abelian, and thus exp is surjective.
Now suppose dimG > 1. Let t be a Cartan subalgebra of g. Then
exp g =
_
gG
exp(Ad(g)t) =
_
gG
c(g) exp t =
_
gG
c(g)T,
where T is the maximal torus with Lie algebra t. So exp g, as the image of the
compact set GT under the continuous map
GT G, (g, t) gtg
1
,
is compact and therefore closed in G. We will show that it is also open, and thus
equals G by connectivity.
Fix any 0 ,= X
0
g, and write g
0
= exp(X
0
). We need to show that exp g
contains a neighborhood of g
0
. Let
A = Z
G
(g
0
)
0
= g G [ g
0
gg
1
0
= g
0
,
the connected component of the centralizer of g
0
in G. Its Lie algebra is given by
a = Z
g
(g
0
) := Y g [ Ad(g
0
)Y = Y .
Note that exp(tX
0
) A for any t R. In particular, g
0
A.
Case 1: dima = dimg. Then a = g, and A = G. So g
0
Z(G). Let t be any
Cartan subalgebra containing X
0
. Then for any X t,
g
0
exp(Ad(g)X) = g
0
c(g)(exp(X)) = c(g)(exp(X
0
) exp(X)) = exp(Ad(g)(X
0
+X)).
Since X is arbitrary and X +X
0
t, we get
g
0
exp(g) exp(g).
115
But exp(g) contains a neighborhood U of e. So exp(g) must also contains g
0
U, a
neighborhood of g
0
.
Case 2: dima < dimg. Since X
0
a, we have 1 dima < n. By induction
assumption, A = exp(a). So
_
gG
g
1
Ag
_
gG
exp(Ad(g)a) exp(g).
We will show that

gG
g
1
Ag contains a neighborhood of g
0
. It follows exp(g)
contains a neighborhood of g
0
.
As before we will take an Ad-invariant inner product on g. Let b = a

. Then
g = ab, and Ad(g
0
) Id is an invertible endomorphism of b. Consider the smooth
map : g = a b G given by
(X, Y ) = g
1
0
exp(Y )g
0
exp(X) exp(Y ).
Then
d(X, 0) =
d
dt
[
t=0
(tX, 0) = X
and
d(0, Y ) =
d
dt
[
t=0
(0, tY ) = Ad(g
0
)Y Y.
It follows that d is an isomorphism at (0, 0). Thus is a dieomorphism from a
neighborhood of 0 g to a neighborhood of e G. Since L
g
1
0
is a dieomorphism,
the set
exp(Y )g
0
exp(X) exp(Y ) [ X a, Y b
contains a neighborhood of g
0
in G. Note that exp(Y ) G, g
0
A, and exp(X) A
for any X a, we conclude that

gG
g
1
Ag contains a neighborhood of g
0
in G.
23.2 Cartans Theorem
We have known that if T is a maximal torus of a compact Lie group G, then
any other maximal torus must have the form gTg
1
for some g G.
Theorem 23.2 (Cartans Theorem). Let G be a compact connected Lie group, T a
maximal torus of G. Then G =

gG
gTg
1
.
Proof. Since exp is surjective,
G = exp(g) =
_
gG
exp(Ad(g)t) =
_
gG
c(g)(exp(t)) =
_
gG
gTg
1
.
116
Remarks. (1) We proved the Cartans theorem using the fact that exp is surjective.
Conversely, if Cartans theorem holds, i.e. every g G is contained in some maximal
torus, then exp is surjective on G since it is surjective on tori. So Cartans theorem
is equivalent to the fact that exp is surjective.
(2) We sketch two other proofs of the Cartans theorem here. For more details,
c.f. D.Bump, Lie Groups, Chapter 16-17.
Geometric proof: On any compact Lie group G there exists a bi-invariant Rie-
mannian metric, under which the geodesics are translations of the one-parameter
subgroups t exp(tX). Now the theorem follow from the fact that any compact
connected Riemannian manifold is geodesically complete.
Topologically proof: The map G/T T G, (g, t) gtg
1
has mapping degree
[W[, where W = N(T)/T is the Weyl group of G with respect to T. In particular,
the map above is surjective.
(3) In general Cartans theorem does not hold for noncompact Lie groups. For
example, consider G = SL(2, R). Then
A =
__
a 0
0 a
1
_
[ a > 0
_
and
N =
__
1 b
0 1
_
[ b R
_
are both maximal connected abelian subgroups of G. Note that any element of A
has trace a + a
1
while any element in N has trace 2. So unless a = 1, an element
of A cannot be conjugate to an element in N.
Corollary 23.3. Let G be a compact connected Lie group and T a maximal torus
of G. Then any g G is conjugate to some t T.
Since any representation is uniquely determined by its character, which is a
conjugate invariant function, we conclude
Corollary 23.4. Let G be a compact connected Lie group, T a maximal torus of G,
and (
1
, V
1
), (
2
, V
2
) are two nite dimensional complex representations of G. Then

1

2
if and only if
1
[
T

2
[
T
.
Remark. Recall that any representation of a torus T can be decomposed into irre-
ducible representations, which are characterized by their weights. So any represen-
tation of G is determined a subset of the weight lattice Z

T
, with multiplicity. This
is called the weight system of the representation. In particular, if we take the repre-
sentation to be the (complexied) adjoint representation Ad C, then any nonzero
vector in the weight system is called a root. The set of roots is called the root sys-
tem of G, which plays an important role in the classication theory of compact Lie
groups.
117
23.3 Applications to Centralizers
Recall that the centralizer of G is
Z(G) = g G [ gh = hg, h G.
Corollary 23.5. If G is a compact connected Lie group, then Z(G) is the intersec-
tion of all maximal tori in G.
Proof. Suppose g Z(G), then for any maximal torus T, there is some h G such
that hgh
1
T. Since hgh
1
= g, g T for any maximal torus T.
Conversely suppose g lies in any maximal torus. For any h G, there is a
maximal torus T such that h T. Since T is abelian, hg = gh. So g Z(G).
In general for any subgroup H G, the centralizer
Z
G
(H) = g G [ gh = hg, h H.
Corollary 23.6. Supposer G is compact, and A G is a connected abelian Lie
subgroup. Then Z
G
(A) is the union of all maximal tori in G that contains A.
Proof. Note that Z
G
(A) = Z
G
(

A). So we may assume A is a torus, otherwise we
may replace A by

A.
Suppose T is a maximal torus containing A. Then by denition T Z
G
(A). So
Z
G
(A) contains the union of all maximal tori in G that contains A.
Conversely, let g Z
G
(A), or equivalently, A Z
G
(g). Then Z
G
(g)
0
is a
compact connected Lie group, and A Z
G
(g)
0
since e A and A is connected. So
one can nd a maximal torus T
1
in Z
G
(g)
0
that contains A. On the other hand,
since exp is surjective, there exists X g such that g = exp(X) Z
G
(g)
0
. So by
denition, g Z(Z
G
(g)
0
). By the previous corollary, g T
1
. In other words, we
nd a torus T
1
Z
G
(g)
0
G that contains both A and g. In particular, if T is any
maximal torus in G that contains T
1
, then g T. This completes the proof.
Corollary 23.7. Supposer G is compact, and A G is a connected abelian Lie
subgroup. Then Z
G
(A) is connected.
Corollary 23.8. Let G be a compact connected Lie group and T a torus in G. Then
T is maximal if and only if Z
G
(T) = T.
Proof. This is an obvious consequence of corollary 3.2.
118
24 The Weyl Groups
24.1 The Weyl Groups
Let G be a compact Lie group, and T G a maximal torus. The normalizer of
T is
N(T) = g G [ gTg
1
= T.
Note that N(T) is a closed subgroup of G, thus also a compact Lie group. By
denition T is a normal subgroup of N(T).
Denition 24.1. The quotient W = N(T)/T is called the Weyl group of G with
respect to T.
Obviously N(gTg
1
) = gN(T)g
1
. So the Weyl groups of G with respect to
dierent maximal tori are isomorphic.
Proposition 24.2. N(T)
0
= T.
Proof. We rst prove that the automorphism group Aut(T) of a torus T = R
k
/Z
k
is isomorphic to GL(k, Z). In particular, it is discrete. To prove this, let : T T
be an automorphism. Then d : R
k
R
k
is an invertible linear map, and we have
the following commutative diagram
R
k

exp
T
k

_
d

R
k

exp
T
k
It follows that d(ker(exp)) ker(exp). In other words, d(Z
k
) Z
k
. So as a k k
matrix, d is actually an integer matrix, i.e. d GL(k, Z). Conversely, any matrix
in GL(k, Z) denes an invertible map that preserves Z
k
, and thus an automorphism
of T.
It follows that any connected group of automorphisms must act trivially. Now
N(T)
0
is a connected Lie group, and the conjugation action of N(T)
0
on T are
automorphisms of T. So N(T)
0
acts trivially on T, i.e. any h N(T)
0
commutes
with all elements in T. So N(T)
0
Z(T) = T. On the other hand, by denition
N(T)
0
T. So N(T)
0
= T.
Corollary 24.3. The Weyl group is a nite group.
Proof. W = N(T)/T is discrete by proposition 1.2. It is compact since N(T) is.
119
Since T is abelian, the conjugation action of T on T itself is trivial. It follows
that the Weyl group acts on T by conjugation.
Proposition 24.4. The conjugation action of W on T is eective.
Proof. This follows from the fact that Z(T) = T since T is maximal.
Proposition 24.5. Let G be a compact connected Lie group, and T a maximal
torus. Then two elements t
1
, t
2
T are conjugate in G if and only if they sit on the
same orbit of the Weyl group action.
Proof. Obviously if w(t
1
) = wt
1
w
1
= t
2
for some w W, then t
1
, t
2
are conjugate
in G.
Conversely if gt
1
g
1
= t
2
. Then gTg
1
gZ
G
(t
1
)g
1
= Z
G
(t
2
). It follows that
both T and gTg
1
are maximal tori in Z
G
(t
2
)
0
. So there exists h Z
G
(t
2
)
0
such
that hgTg
1
h
1
= T. It follows that hg N(T). Moreover, hg(t
1
) = hgt
1
g
1
h
1
=
ht
2
h
1
= t
2
. This completes the proof.
As a consequence, all class (i.e. conjugate invariant) functions on G is in one-
to-one correspondence to W-invariant functions on T.
24.2 The Maximal Tori of Classical Groups
24.2.1 G = U(n) or SU(n)
Recall that
T =
_

_
_
_
_
e
it
1
.
.
.
e
itn
_
_
_
: t
i
R
_

_
is a maximal torus in U(n).
Similarly one can study G = SU(n) = U(n) SL(n, C). Then a maximal torus
is given by

T =
_

_
_
_
_
e
it
1
.
.
.
e
itn
_
_
_
: t
i
R, t
1
+ +t
n
= 0
_

_
.
24.2.2 G = SO(n)
120
Next lets study the special orthogonal group SO(n), which is the identity com-
ponent of O(n). We rst assume n = 2l is even. Then
T = SO(2) SO(2) =
_

_
_
_
_
_
_
_
_
cos t
1
sin t
1
sin t
1
cos t
1
.
.
.
cos t
l
sin t
l
sin t
l
cos t
l
_
_
_
_
_
_
_
: t
i
R
_

_
is a torus in SO(n). Moreover, from linear algebra we know that any orthogonal
matrix is conjugate using orthogonal matrices to a matrix in T. In other words,
SO(n) =
gSO(n)
gTg
1
. According to the next lemma, T is a maximal torus in
O(n).
Lemma 24.6. Suppose G is a connected compact Lie group, T a torus in G. Then
T is maximal if and only if G = gTg
1
.
Proof. If T is maximal, this is the Cartans theorem.
Now suppose G = gTg
1
. Then the following map
G/Z(T) T G, (gZ(T), t) gtg
1
is surjective, i.e. dimG/Z(T) +dimT dimG. So dimZ(T) dimT. This implies
T is maximal, since if T is not maximal, then Z(T) contains all maximal torus that
contains T, and thus dimZ(T) > dimT. This completes the proof.
Similarly for n = 2l + 1 an odd number, a maximal torus of SO(n) is
T = SO(2) SO(2)1 =
_

_
_
_
_
_
_
_
_
_
_
cos t
1
sin t
1
sin t
1
cos t
1
.
.
.
cos t
l
sin t
l
sin t
l
cos t
l
1
_
_
_
_
_
_
_
_
_
: t
i
R
_

_
24.2.3 G = Sp(n)
Now consider the symplectic group Sp(n) = Sp(2n, C) U(2n). Sp(n) consists
of unitary matrices of the form
_
A

B
B

A
_
. There is a canonical inclusion
U(n) Sp(n), A
_
A 0
0

A
_
.
121
We denote by T the image of the maximal torus of U(n) described above under this
inclusion.
Proposition 24.7. T is a maximal torus in Sp(n).
Proof. Obviously T is a torus in Sp(n). We will show Z(T) = T. Suppose X =
_
A

B
B

A
_
Z(T). Then
_
iI
n
0
0 iI
n
__
A

B
B

A
_
=
_
A

B
B

A
__
iI
n
0
0 iI
n
_
implies B = 0 and thus A U(n). Now the fact X Z(T) implies that A commutes
with all diagonal unitary matrices. It follows that A is diagonal itself, and thus
X T.
24.3 The Weyl Groups and Root Systems of Classical Groups
We only do this in detail for U(n). To compute the Weyl group of U(n), we rst
notice that if gt
1
g
1
= t
2
for t
1
, t
2
T, then t
1
and t
2
have the same eigenvalues. In
other words, as diagonal matrices the entries of t
2
are permutations of entries of t
1
.
It follows that the Weyl group acts on a generic element t = diag(e
it
1
, , e
itn
) by
permuting t
i
s. So W is a subgroup of the full symmetric group S(n). On the other
hand, since
_
0 e
i
e
i
0
__
x 0
0 y
__
0 e
i
e
i
0
_
=
_
y 0
0 x
_
.
we see that any monomial matrix (matrices with a single nonzero entry in each row
and column) in U(n) is in the normalizer of T. It follows that N(T)/T S
n
. So
the Weyl group of U(n) is W(U(n)) = S(n).
We can also compute the root system of U(n). Recall that the root system
consists of the nonzero weights for the complexied adjoint representation restricted
to T. Also recall the following facts:
The Lie algebra of U(n) is the space of n n skew-Hermitian matrices.
A matrix X is skew Hermitian if and only if iX is Hermitian.
Any matrix in gl(n, C) can be written uniquely as the sum of a Hermitian
matrix and a skew-Hermitian matrix.
It follows that the complexied Lie group is gl(n, C). The complexied coadjoint
action of U(n) on gl(n, C) is still given by
Ad(g)(X) = gXg
1
.
122
Now take g = diag(e
it
1
, , e
itn
), then
Ad(g)(x
jk
) = g(x
jk
)g
1
= (e
i(t
j
t
k
)
x
jk
).
So if we take E
jk
to be the matrix with 1 at the jk entry and 0 elsewhere, then
C E
jk
is a one-dimensional (and thus irreducible) subrepresentation, with weight
t
j
t
k
(we think t
j
as the function on the tangent space of T that picks the jth
coordinate function). As a consequence, the root system for U(n) is
(U(n)) = t
j
t
k
[ j ,= k.
The Weyl group of SU(n) is still S(n), the full symmetric group, and its root
system is still
(SU(n)) = t
j
t
k
[ j ,= k.
(However, in this case we have the restriction t
1
+ +t
n
= 0.)
We end this lecture by listing the Weyl groups and root systems for other classical
groups. For details, c.f. T.Brocker & T.Dieck.
The Weyl group of SO(2l + 1) is G(l), the group of permutations of the
set l, , 1, 1, , l with (k) = (k) for all 1 k l. The root
system for SO(2l + 1) is
(SO(2l + 1)) = t
j
t
k
[ 1 j < k l t
j
[ 1 j l.
The Weyl group of SO(2l) is the subgroup SG(l) of G(l) that consists of even
permutations. The root system for SO(2l) is
(SO(2l)) = t
j
t
k
[ 1 j < k l.
The Weyl group of Sp(n) is still G(n). The root system of Sp(n) is
(Sp(n)) = t
j
t
k
[ 1 j < k n 2t
j
[ 1 j n.
123
25 Structure of Compact Lie Groups
25.1 Structure of Compact Lie Algebras
Let g be a Lie algebra. Recall that a subset h g is an ideal if [g, h] h. In
particular any ideal is a Lie subalgebra.
Example: (1) The center Z(g) is an ideal because [g, Z(g)] = 0.
(2) The derived Lie subalgebra g

of g is the subalgebra spanned by all elements

in [g, g]. It is obviously an ideal.
Denition 25.1. Let g be a Lie algebra.
1. g is simple if it has no nonzero proper ideals and if dimg > 1.
2. g is semisimple if it is a direct sum of simple Lie algebras.
3. g is reductive if it is a direct sum of a semisimple Lie algebra and an abelian
Lie algebra.
Our rst theorem shows that the Lie algebra of a compact Lie group is reductive.
Theorem 25.2. Let G be a compact Lie group, then g is reductive. More explicitly,
g = g

Z(g),
where Z(g) is the center of g, which is abelian, and g

is semisimple.
Proof. Choose any adjoint invariant inner product on g. Then we have seen in
lecture 22 that for any X g, ad(X) is skew-symmetric. It follows that if a is an
ideal of g, so is a

. As a consequence, g can be decomposed into a direct sum of

minimal ideals
g = s
1
s
k
z
1
z
l
,
where dims
j
> 1 and dimz
j
= 1. Obviously
s = s
1
s
k
is semisimple, and
z = z
1
z
l
is abelian. It remains to show s = g

and z = Z(g).
First notice that [s
i
, z
j
] s
i
z
j
= 0 for any i, j. So [s
i
, z
j
] = 0 for any i, j.
Similarly [z
i
, z
j
] = 0 for all i ,= j. Moreover, [z
i
, z
i
] = 0 for any i since dimz
i
= 1.
124
So [g, z
i
] = 0 for all i. In other words, z = z
1
z
l
Z(g). Conversely, suppose
Z Z(g). Decompose Z as
Z = S
1
+ +S
k
+Z
1
+ +Z
l
,
where S
i
s
i
, and Z
i
z
i
. Then 0 = [Z, s
i
] = [S
i
, s
i
], i.e., S
i
Z(s
i
). Since s
i
is a
minimal ideal of dimension dims
i
> 1, we must have S
i
= 0. So Z z. This proves
z = Z(g).
Similarly for i ,= j, [s
i
, s
j
] = 0. We claim that for any i, s

i
= span[s
i
, s
i
] = s
i
. In
fact, since dims
i
> 1 and s
i
Z(g) = 0, dims

i
1. So dims

i
= dims
i
, otherwise
it is a nonzero proper ideal. Now it follows that
span[g, g] = span[s
1
, s
1
] span[s
k
, s
k
] = s
1
s
k
,
or in other words, s = g

.
25.2 The Commutator Subgroup
Recall that for any group G, its commutator subgroup G

is the normal subgroup

generated by elements of the form g
1
g
2
g
1
1
g
1
2
.
Theorem 25.3. Let G be a compact connected Lie group. Then G

is a connected
closed normal Lie subgroup of G with Lie algebra g

.
Proof. Since G is compact, it is a closed Lie subgroup of U(N) for some N. Decom-
pose the standard representation of G on C
N
into irreducible ones,
C
N
= C
n
1
C
n
k
,
where n
1
+ + n
k
= N, and denote by
i
the irreducible representation of G on
C
n
i
.
Consider the map
: G (S
1
)
k
, g (det
1
(g), , det
k
(g)).
Then this is a Lie group homomorphism. It follows that H = ker() is a closed
Lie subgroup of G with Lie algebra h = ker(d). We will show that h = g

, and
H
0
= G

, which will nish the proof.

We rst show h = g

. Suppose Z Z(g), then e

tZ
Z(G). By Schurs lemma,

i
(e
tZ
) = c
i
(t)Id for some scalar c
i
(t) with c
i
(0) = 1. So if we think of G as a closed
subgroup of U(n
1
) U(n
k
), then Z is given by the diagonal matrix
d
dt
[
t=0
(
1

k
)(e
tZ
) = diag(c

1
(0), , c

1
(0), , c

k
(0), , c

k
(0)).
125
It follows that
d(Z) = (d det
d
dt
[
t=0

1
(e
tZ
), , d det
d
dt
[
t=0

k
(e
tZ
)) = (n
1
c

1
(0), , n
k
c

k
(0)).
As a consequence, ker(d) Z(g) = 0. On the other hand, since tr(AB) =
tr(BA), for any X g

we must have tr(d

i
(X)) = 0. So det
i
(e
tX
) 1. We thus
get d(X) = 0 for all X g

, i.e. g

ker(d). Combine this with ker(d)Z(g) =

0, we conclude ker(d) = g

.
Finally we show H
0
= G

. Obviously G

H since determinant is multiplicative.

G

is connected since G

=
j
U
j
, where U = g
1
g
2
g
1
1
g
1
2
[ g
1
, g
2
G is connected,
and e U
j
for all j. So G

H
0
. It remains to show H
0
G

, or equivalently,
to show G

contains a neighborhood of e in H. Recall from lecture 9 that for any

X, Y h, [X, Y ] is the tangent vector of the curve
t c
X,Y
(t) = exp(

tX) exp(

tY ) exp(

tX) exp(

tY )
at t = 0. Now let [X
1
, Y
1
], , [X
m
, Y
m
] be a basis of g

. Consider the map

c : R
m
H, (t
1
, , t
m
) c
X
1
,Y
1
(t
1
) c
Xm,Ym
(t
m
).
Then dc(0) is an isomorphism onto h. So c is locally a dieomorphism near 0. Thus
the image of c contains a neighborhood of e in H. This completes the proof since
the image of c is contained in G

.
Corollary 25.4. G

Z(G) is a nite group.

Proof. We have just seen in the proof that any g Z(G), (g) must be the diagonal
matrix diag(c
1
, , c
1
, , c
k
, , c
k
). If we also have g G

= ker(), then
c
n
1
1
= = c
n
k
k
= 1. So c
i
is an n
th
i
-root of unity. It follows that G

Z(G) is a
nite group.
Proposition 25.5. Let g

= s
1
s
k
be the decomposition of g

into simple
ideals and let S
i
= exp s
i
. Then
(1) S
i
is a connected closed Lie subgroup of G

with Lie algebra s

i
.
(2) Any proper closed normal Lie subgroup of S
i
are nite and lies in the center
of S
i
.
Proof. (1) Let K
i
= g G

[ Ad(g)[
s
j
= Id, j ,= i
0
. Then K
i
is a connected
closed Lie subgroup of G

. We will show K
i
= S
i
. In fact,
X k
i
exp(tX) K
i
, t R
exp(tad(X))[
s
j
= Ad(exp(tX))[
s
j
= Id, j ,= i
ad(X)(s
j
) = 0, j ,= i
[X, s
j
] = 0, j ,= i.
126
So s
i
k
i
. Conversely, if X k
i
, then projection of X onto s
j
, j ,= i, must be zero.
So k
i
= s
i
. It follows K
i
= exp(k
i
) = exp(s
i
) = S
i
.
(2) Suppose N is a proper normal Lie subgroup of S
i
, i.e. sNs
1
= N for all
s S
i
. Taking derivative, we get Ad(s)n = n for any s S
i
. Taking derivative
again, we have ad(X)n n for any X s
i
. So n is an ideal of s
i
. It follows that
n = 0. So N is discrete. Since N is closed, it is also compact. So N is nite.
To show N is abelian, for each n N, consider
C
n
= sns
1
[ s S
i
.
It is connected since S
i
is connected. Since N is normal, C
n
N. So C
n
contains
only one element since N is discrete. It follows that C
n
= n, and thus n lies in
the center of S
i
.
25.3 Structure of Compact Lie Groups
Theorem 25.6. Let G be a compact connected Lie group. Then
1. G is the product of a semisimple Lie group with an abelian Lie group,
G = G

Z(G)
0
.
2. Let F = (g, g
1
) [ g G

Z(G)
0
, then F is nite and
G

= (G

Z(G)
0
)/F.
3. There is a nite abelian subgroup F

of S
1
S
k
such that
G

= (S
1
S
k
)/F

.
Proof. (1) Since G

is closed, it is compact. So G

= exp g

, and
G = exp g = exp(g

Z(g)) = G

Z(G)
0
.
(2) F is nite since G

Z(G)
0
G

Z(G) and the later is nite. From (1),

the Lie group homomorphism
G

Z(G)
0
G, (g, z) gz
is surjective. The kernel of this map is F. So G

= (G

Z(G)
0
)/F.
(3) Since [s
i
, s
j
] = 0 for i ,= j,
G

= exp g

= exp(s
1
s
k
) = S
1
S
k
.
So the Lie group homomorphism
S
1
S
k
G

, (s
1
, , s
k
) s
1
s
k
is surjective. Moreover, its dierential at e is the identity map. So the kernel of this
map is a discrete normal subgroup, which has to be nite and lies in the center.
127
26 The Weyl Integration Formula
26.1 The quotient G/T
Suppose G is a compact Lie group, and T G a maximal torus. We have known
from previous lectures that
There exist normalized Haar measure dg, dt on G, T respectively.
The quotient G/T is a homogeneous G-manifold with tangent space
T
eT
(G/H) = g/t := p.
Consider the restriction of the adjoint action of G on g to T. Obviously any
t t is xed by Ad(T)-action. Conversely,
Lemma 26.1. Any vector in g xed by Ad(T) is in t.
Proof. Suppose X g is xed by Ad(T), but X , t. Then exp(tX) is a one-
parameter subgroup not contained in T which commutes with T. It follows that the
subgroup T

generated by T and exp(tX) is abelian. So T is not maximal.

Since p contains no Ad(T)-invariant vectors, we have
Corollary 26.2. dimGdimT is even.
Proof. p is a real representation of T which contains no Ad(T)-invariant vector,
its irreducible components must be of dimension 2. (Since any irreducible real
representation of a torus must be either trivial or of dimension 2.)
We will x an adjoint invariant inner product on g, and identify p with the
orthogonal complement of t in g,
g = t p.
So in particular p g.
We will also choose the quotient density
d(gT) = dg/dt
on the quotient G/T. This is also a G-invariant normalized density, and for any
f C(G),
_
f(g)dg =
_
G/T
_
T
f(gt)dtd(gT).
128
(Note that for any f C(G), the function
g
_
T
f(gt)dt
is constant on the orbit gT, and thus denes a function on G/T.)
Remark. The arguments above apply to G/H for any closed subgroup H.
26.2 The Weyl Integration Formula
Theorem 26.3 (Weyl Integration Formula for Class Functions). Suppose G is com-
pact, and f a class function on G. Denote by dg and dt the Haar measures on G
and T respectively. Then
_
G
f(g)dg =
1
[W[
_
T
f(t) det([Ad(t
1
) I
p
][
p
)dt
Proof. Consider the map
: G/T T G, (gT, t) gtg
1
.
We will compute the Jacobian J of . The Jacobian of at (gT, t) is the same as
the Jacobian of
R(t
1
) c(g
1
) (L(g) L(t))(hT, s) = htsh
1
t
1
at (eT, e) since both L(g) R(t) on G/T T and R(t
1
) c(g
1
) on G preserves the
given densities on these spaces. The tangent map of the latter at (eT, e) is given by
A : (X + t, S) (Id Ad(t))(X) +Ad(t)S.
It follows that the Jacobian of equal the determinant of the pervious map, i.e.
(J)(gT, t) = det([Ad(t
1
) Id][
p
) det Ad(t) = det([Ad(t
1
) Id][
p
),
where we used the fact that det Ad(t) = 1 for any compact Lie group, since compact
Lie groups are unimodular.
It follows that is a locally a dieomorphism over a dense open set. Moreover,
(g
1
T, t
1
) = (g
2
T, t
2
) if and only if t
1
, t
2
T conjugate in G, or equivalently, lie
in the same W-orbit. So is a [W[-to-one covering map over a dense open set. It
follows that for any class function f,
_
G
f(g)dg =
1
[W[
_
G/TT
f((gT, t))J()dgdt
=
1
[W[
_
T
f(t) det([Ad(t
1
) I
p
][
p
)dt.
129
Note that for any continuous function f C(G), the function

f(t) =
_
G
f(gtg
1
)dg
is a W-invariant function on T, which can be identied with a class function on G.
Moreover,
_
G
f(g)dg =
_
G

f(g)dg.
So we have
Corollary 26.4 (Weyl Integration Formula for Continuous Functions). For any
continuous function f on G,
_
G
f(g)dg =
1
[W[
_
T
det([Ad(t
1
) I
p
][
p
)(
_
G
f(gtg
1
)dg)dt.
26.2.1 Example: U(n)
Now consider G = U(n) the unitary group. Let T be the maximal torus consists
of all diagonal matrices in U(n). In other words, any t T has the form
t =
_
_
_
e
it
1
.
.
.
e
itn
_
_
_
.
Let dt be the normalized Haar measure on T.
Proposition 26.5. In this setting, det([Ad(t
1
) I
p
][
p
) =

j<k
[e
it
j
e
it
k
[
2
.
Proof. We may think of Ad(t
1
) I
p
as a linear transformation on the complexied
vector space pC. Recall that for G = U(n), the complexied Lie algebra u(n)C =
gl(n, C). So p C is the vector subspace spanned by the T-eigenspaces in gl(n, C)
corresponding to nontrivial characters of T. These are spanned by the elementary
matrices E
jk
, j ,= k, which are the matrices with the only nonzero entry a 1 at
the (j, k)-position. Since the eigenvalues of T on E
jk
is e
it
j
e
it
k
, we get
det([Ad(t
1
) I
p
][
p
) =

j=k
(e
it
j
e
it
k
1)
=

j<k
(e
it
j
e
it
k
1)(e
it
k
e
it
j
1)
=

j<k
(e
it
j
e
it
k
)(e
it
j
e
it
k
)
=

j<k
[e
it
j
e
it
k
[
2
.
130
It follows that for any class function f on U(n),
_
U(n)
f(g)dg =
1
n!
_
T
f(t)

j<k
[e
it
j
e
it
k
[
2
dt.
131